bell labs transmission cohérentesur fibre optique
TRANSCRIPT
1 © Nokia 2016
Bell Labs
Transmission cohérente sur fibreoptiqueCFOR: Technologies des réseaux à fibres optiques multi-terabit/s
• Gabriel CHARLET
• 1er Dec. 2016
<Change information classification in footer>
2 © Nokia 2016
Introduction
• Long haul optical systems transport almost all data traffic (>99%).
• Undersea optical systems are the successor of the first telegraphic undersea cables !
• Transport data, video, internet, voice (fix & mobile) over a single network.
• Initial deployment of optical networks in the 1980s.
3 © Nokia 2016
• Direct detection
• Basics of coherent detection
• Implementation of coherent transponder
• System performances obtained thanks to coherent
• Conclusion
OUTLINE
4 © Nokia 2016
• Benefits from fiber optics:
- Low attenuation ~0.2dB/km
- Wide bandwidth with low attenuation (>>10THz)
- Efficient optical amplification (Erbium doped fiber amplifier over 2 x 5THz)
<Change information classification in footer>
Optical data transmission
Encoding bits on an optical signal, fiber transmission and reverse operation
Fiber optics
RX
0100101110110101 0100101110110101 electronic
optic
TX
5 © Nokia 2016
Point to point WDM system
TX1 λ1
TX2 λ2
TX λ80
RX1 λ1
RX2 λ2
RX λ80
Mux
Fiber
Demux
Optical amplifier
• Typically 80 to 100 channels (wavelengths) with 50GHz spacing
• 100-200Gb/s per channels typical in 2016
• Span length : 50 to 100km (10 to 30dB span loss typically)
• 10-20 amplifiers for terrestrial systems, up to 200 for undersea systems !
6 © Nokia 2016 <Change information classification in footer>
Electrical field and spectrum of laser
f
Constellation diagram SpectrumEvolution of optical field
Ex
0 1 Re
~5fs (@1.55µm)
Im
t
• 1.55µm used in long distance telecom industry to benefit from the lowest attenuation of fiber and of optical amplification from EDFA.
• This wavelength corresponds to a frequency of ~200THz
Laser
7 © Nokia 2016 <Change information classification in footer>
Modulation format, constellation diagram, spectrum
2 x B
Constellation diagram SpectrumEvolution of optical field
Ex 10 011
0 1
OOK
Re
25ps (@40Gb/s) ~5fs (@1.55µm)
Im
t
• Constellation diagram represents the field of the electrical signal at the center of eachsymbol.
• Spectrum width (of first lobe) is twice the modulation speed
Laser Modulator
B
8 © Nokia 2016
Direct detection
Direct detection is the conventional way to detect optical signal.
= QUADRATIC detection of electric field,
Optical signalElectrical signal~|As(t)|²
Ex 10 011
25ps (@40Gb/s) ~5fs (@1.55µm)
t
Decision element
1 0 1 1 0
9 © Nokia 2016
• Direct detection
• Basics of coherent detection
• Implementation of coherent transponder
• System performances obtained thanks to coherent
• Conclusion
OUTLINE
10 © Nokia 2016 <Change information classification in footer>
Modulation format and constellation diagramPhase shift keying
2 x B
Constellation diagram SpectrumEvolution of optical field
Ex 10 011
0 1
OOK
Re
25ps (@40Gb/s) ~5fs (@1.55µm)
2 x B
Ex 0 π 0 0 π BPSK0π
Re
Im
Im
QPSK5π/4 3π/4π/4Ex
5π/4
3π/4
−π/4
π/4
2 x B/2Re
Im
t
OOK : on off keyingBPSK : binary phase shift keyingQPSK : quadrature phase shift keying
11 © Nokia 2016
Differential detection
Direct detection is the conventional way to detect optical signal.
= QUADRATIC detection of electric field,
Differential detection= Optical demodulator + direct detection
- Suited for detection of phase modulated signal
Optical signalElectrical signal
Optical signalElectrical signal
~|As(t)|²
~|As(t)+As(t+T)|²
As(t)
As(t+T)
12 © Nokia 2016
Coherent detection= LINEAR detection of the electric field, by beating with local oscillator (LO)
- LO frequency = Signal frequency � homodyne detection
- LO frequency ≠ Signal frequency � heterodyne detection
- LO frequency ≈ Signal frequency � intradyne detection
<Change information classification in footer>
Coherent detection
Optical signal
Local oscillator (LO)
Electrical signal
~As.ALO
13 © Nokia 2016
• No optical amplifier available at that time.
• Transmission distance was mostly limited by fiber attenuation.
• « Amplification » provided by strong local oscillator.
• Amplitude shift Keying (ASK) vs Phase shift Keying (PSK) ?
<Change information classification in footer>
Motivation for coherent detection in the 1980’s
150km fiber 30dB
150km fiber 75km 45dB
As.ALO vs As.As
14 © Nokia 2016 <Change information classification in footer>
Coherent detection in the 1980’s
Homodyne or heterodyne ?
f
Modulatedsignal
BW
LO
15 © Nokia 2016 <Change information classification in footer>
Coherent detection in the 1980’s
Homodyne or heterodyne ?
But EDFA was invented…
f
Modulatedsignal
BW
LO
16 © Nokia 2016
• Technological evolutions:
- Availability of high speed ADC (high bandwidth, high sampling rate)
- Strong progress of CMOS digital signal processing capabilities
• Optical transmission landscape
- Phase shift keying modulation used again! (with differential detection) DPSK, DQPSK
- Electronic Signal processing (FFE, DFE,… for chromatic dispersion, PMD… mitigation)
- Willingness to increase spectral efficiency
<Change information classification in footer>
Renewal interest for coherent detection ~2005
First “modern” coherent detection experiment in 2004-2005
17 © Nokia 2016
Coherent detection= LINEAR detection of the electric field, by beating with local oscillator (LO)
- LO frequency = Signal frequency � homodyne detection
- LO frequency ≠ Signal frequency � heterodyne detection
- LO frequency ≈ Signal frequency � intradyne detection
<Change information classification in footer>
Intradyne coherent detection
Optical signal
Local oscillator (LO)
Electrical signal
~As.ALO
18 © Nokia 2016 <Change information classification in footer>
Interference at coherent mixer outputCoherent mixer or 90° hybrid
))(( ttis
seA ϕω +
Coherent Mixer
)]()sin[(4)(2 ttAAtI loslos ϕωω +−=
)]()cos[(4)(1 ttAAtI loslos ϕωω +−= In phase “I”
Quadrature “Q”
tilo
loeA ω(1)
(2)
(1)+(2)
(1)-(2)
(1)+j(2)
(1)-j(2)Local oscillator
Signal
)()()( 432 tPDtPDtI −=
)()()( 211 tPDtPDtI −=+-
+-
π
0
π/2
−π/2
)]()cos[(2)).((22*
1 ttAAAAEEEEPD olsolsolsolsols ϕωω +−++=++=
)]()cos[(2)).((22*
2 ttAAAAEEEEPD olsolsolsloslos ϕωω +−−+=−−=
19 © Nokia 2016
Polarization diversity receiversignal polarization fluctuate over time => polarization diversity RX required
0100011…
Q//Pol. splitter
Local oscillator
ADC
ADC
ADC
ADC
DS
P
I
Q
I//signal
Coherentmixer
3dB
//
Coherentmixer
20 © Nokia 2016 <Change information classification in footer>
Polarization division multiplexing: PDM
Double the bit rate transported
QPSK5π/4 3π/4π/4Ex
5π/4
3π/4
-π/4
π/4
2 x B/2Re
Im
QPSK5π/4 3π/43π/4Ey
5π/4
3π/4
-π/4
π/4
2 x B/2Re
Im
PDM QPSK
5π/4
3π/4
-π/4
π/4
Re
Im
5π/4
3π/4
-π/4
π/4
Re
21 © Nokia 2016
• Direct detection
• Basics of coherent detection
• Implementation of coherent transponder
• System performances obtained thanks to coherent
• Conclusion
OUTLINE
22 © Nokia 2016
Digital coherent receiver versus direct-detected systems…
10 Gb/s NRZ receiver(direct-detection)
Coherent receiver
signal
Singly-polarized
signal
Local oscillator
Coherentmixer
signal
ADC
Polarization-multiplexed signal
X
Y
arbitrary state of polarisation
(SOP)
Analog-to-Digital
Converters
Digital Signal
Processing
BER
Photodiode
Decisiongate
Amplitude, phase and
polarization!!
just intensity…
X
Y
DSP
23 © Nokia 2016 <Change information classification in footer>
Coherent receiver
Digital signal processing
CoherentMixer
Local oscillator
Sampling(scope)
DSP(PC)
Fiber out
Storage(scope)
ADC
ADC
ADC
ADC
j
j
(I+jQ)//
(I+jQ)
Pol
ariz
atio
n D
emul
tiple
and
Equ
aliz
atio
n
Car
rier
freq
uenc
yan
dP
hase
Est
imat
ion
Sym
bol
iden
tific
atio
n
BE
R &
Q²-
fact
or
stor
age
CD
com
pens
atio
n
+
+
scope
100Gb/s PDM-QPSK (experimental data)
24 © Nokia 2016 <Change information classification in footer>
Coherent receiver
Digital signal processing
CoherentMixer
Local oscillator
Sampling(scope)
DSP(PC)
Fiber out
Storage(scope)
ADC
ADC
ADC
ADC
j
j
(I+jQ)//
(I+jQ)
Pol
ariz
atio
n D
emul
tiple
and
Equ
aliz
atio
n
Car
rier
freq
uenc
yan
dP
hase
Est
imat
ion
Sym
bol
iden
tific
atio
n
BE
R &
Q²-
fact
or
stor
age
CD
com
pens
atio
n
+
+
scope
100Gb/s PDM-QPSK (experimental data)
25 © Nokia 2016 <Change information classification in footer>
Coherent receiver
Digital signal processing
CoherentMixer
Local oscillator
Sampling(scope)
DSP(PC)
Fiber out
Storage(scope)
ADC
ADC
ADC
ADC
j
j
(I+jQ)//
(I+jQ)
Pol
ariz
atio
n D
emul
tiple
and
Equ
aliz
atio
n
Car
rier
freq
uenc
yan
dP
hase
Est
imat
ion
Sym
bol
iden
tific
atio
n
BE
R &
Q²-
fact
or
stor
age
CD
com
pens
atio
n
+
+
scope
100Gb/s PDM-QPSK (experimental data)
26 © Nokia 2016 <Change information classification in footer>
Coherent receiver
Digital signal processing
CoherentMixer
Local oscillator
Sampling(scope)
DSP(PC)
Fiber out
Storage(scope)
ADC
ADC
ADC
ADC
j
j
(I+jQ)//
(I+jQ)
Pol
ariz
atio
n D
emul
tiple
and
Equ
aliz
atio
n
Car
rier
freq
uenc
yan
dP
hase
Est
imat
ion
Sym
bol
iden
tific
atio
n
BE
R &
Q²-
fact
or
stor
age
CD
com
pens
atio
n
+
+
scope
100Gb/s PDM-QPSK (experimental data)
27 © Nokia 2016 <Change information classification in footer>
Coherent receiver
Digital signal processing
CoherentMixer
Local oscillator
Sampling(scope)
DSP(PC)
Fiber out
Storage(scope)
ADC
ADC
ADC
ADC
j
j
(I+jQ)//
(I+jQ)
Pol
ariz
atio
n D
emul
tiple
and
Equ
aliz
atio
n
Car
rier
freq
uenc
yan
dP
hase
Est
imat
ion
Sym
bol
iden
tific
atio
n
FE
C
stor
age
CD
com
pens
atio
n
+
+
scope
100Gb/s PDM-QPSK (experimental data)
28 © Nokia 2016
• Each symbol has a given spectral extension (~Baudrate).
<Change information classification in footer>
Chromatic dispersion compensation block
t
f
S1 S2 S3 S4 S5 S6 S7 S8
29 © Nokia 2016
• Each symbol has a given spectral extension.
• After transmission and chromatic dispersion impact, symbols overlap each others
• Frequency domain implementation to delay some frequencies and resynchronize all frequency components from transmitter symbol.
• Each symbol can overlap > 1000 symbols for transoceanic links.
<Change information classification in footer>
Chromatic dispersion compensation block
t
f
30 © Nokia 2016
• 2010: first apparition of high speed CMOS ADC
• 56GS/s sampling rate, 8 bit resolution and ~15GHz analog bandwidth
• Total amount of data to be processed for 100Gb/s PDM-QPSK: 4x56x8= 1.8Tb/s!
• Unpractical to exchange 1.8Tb/s between 2 separate chips => single chip required for ADC and DSP. CMOS required for low power DSP.
• In 2010, first ADC-DSP chip with 65nm CMOS technology including
• 4 ADCs (56GS/s, 8bits)
• CD compensation (~2000km)
• Clock recovery
• MIMO 2x2 (polarization deMux, equalization)
• Optical carrier frequency estimation
• Optical carrier phase estimation
• FEC (product code BCHxBCH)
<Change information classification in footer>
ADC and DSP in a single CMOS chip
31 © Nokia 2016
MIMO processing (polarization demultiplex. & equalization)
• Equalization is NOT done thanks to channel estimation (through training sequence or pilot), NOR by using decision on the symbol after full DSP processing.
• Blind equalization using only the intensity information (before phase processing). => no overhead associated with training sequence or pilot.
+
+
feedback
xout
FIR filter (N taps)hxx
hyy
hyx
hxy
Butterfly structure
yout
xin
yin
32 © Nokia 2016 <Change information classification in footer>
Polarization Mode Dispersion (PMD) and coherent detection
• PMD caused by fiber optics imperfections (stress, non perfect circularsymmetry…).
• PMD is a stochastic effect. Differential group delay (DGD) can be ~3 times larger than average DGD value.
• PMD has been a major limitation for the deployment of 40Gb/s systems.
• But entirely solved with MIMO equalizer of coherent receiver !
z
x
y
z0
zL
t
x
y
à z0
t
x
y
à zL
DGD (ps)
33 © Nokia 2016
• Phase estimation is a key part allowing to avoid the use of « optical phase locked laser » of the 1980’s.
• Preceded by coarse frequency estimation to allow for several GHz of frequencyoffset between TX and RX laser.
• Various algorithms available, here for QPSK.
<Change information classification in footer>
Carrier phase estimation (CPE)Required because intradyne detection is used
∑( · )N(.)4 arg( · )/4 unwrap
Ca
rte
sia
n=
> P
ola
r
delay
Po
lar=
>
Ca
rtesia
n
Removedata
Average noise
φ+κπ/2+Ν4φ+4Ν 4φ φ
+π/4
34 © Nokia 2016
• Optical communication requested BER is 10-13 (or even 10-16).
• Maximization of transmission reach and/or capacity is obtained through the use of forward error correction similarly to any other digital systems.
• But processing of up to 500Gb/s in a single chip !
• Code rate (r) vs overhead (OH):
• Code rate r=0.8 => overhead 25%
• Code rate r=0.5 => overhead 100%
• 2000: 7% overhead and code RS 255-239: FEC threshold 10-4 (Q~11.5dB)
• 2003: 7% overhead and code BCHxBCH: FEC threshold 4 10-3 (Q~8.5dB)
• 2012: 25% overhead and “soft decision” LDPC (low density parity check): FEC threshold ~2 10-2 (Q~5.5dB)
<Change information classification in footer>
Bit error rate and forward error correction (FEC)
)1/1( −= rOH
K bitsFEC encoding
K bits Overhead
N bits
NKr /=Code rate
35 © Nokia 2016
100
1000
10000
2010 2014 2018 2022 2026
0
20
40
60
80
100
120
2010 2014 2018 2022 2026
<Change information classification in footer> © Nokia 2016
Maximizing bit rate per device to minimize cost per bitMaximizing baudrate, format level & carrier count per ASIC
Ba
ud
rate
(Gb
au
d)
Bit
rate
(Gb
it/s
)
200G
1T
65
nm
40
nm 28
nm
Tremendous progress of high speed optical coherent interfaces
100G
Dual λ400G
X2 for each generation
36 © Nokia 2016
multi-channel photonic integration:
• Dual PDM I/Q modulator
• dual tunable laser
• dual coherent receiver
• dual ADC/DSP/DAC chip…
<Change information classification in footer>
Opto-electronic components
PDM I/Q modulator (LiNbO3, InP, siPho)
Quad driver (InP, GaAs)
Coherent receiver (InP, SiPho, free space…)
ADC, DSP, DAC (CMOS)
Tunable laser (InP)
37 © Nokia 2016
Spectrum shaping thanks to DAC (Digital to Analog Convertor) at TX
Spectrum shaping thanks to digital filter in transmitter to reduce spectrum occupancy and increase spectrum efficiency
Time domain(impulse response)
Frequency domain(spectrum)
t
Sinc
T
f
B=1/T
Rect
f t
T2B
Log
scal
eAmplitude
Wavelength (0.1nm/div)
Pow
er (
5dB
/div
)
“NRZ”RRC 0.01
2B
1.01B
38 © Nokia 2016
• Direct detection
• Basics of coherent detection
• Implementation of coherent transponder
• System performances obtained thanks to coherent
• Conclusion
OUTLINE
39 © Nokia 2016
1
2
4
8
16
32
64
2000 2005 2010 2015
<Change information classification in footer> © Nokia 2016
Progress in submarine capacity per fiber(transmission distance > 6000km)
Ca
pa
city
(Tb
/s)
10G
40G
100G
200G+
400G
Strong capacity increase thanks to coherent detection
Direct/diffdetection
coherentdetection
40 © Nokia 2016 <Change information classification in footer>
Shannon limitFor dual polarization signal, AWGN channel
0
2
4
6
8
10
12
14
-2 2 6 10 14 18 22
Info
rma
tio
n b
its
pe
r sy
mb
ol
SNR [dB]
QPSK
16QAM
64QAM
C=2. log2(1+SNR)
41 © Nokia 2016 <Change information classification in footer>
Example of recent record experiments
0
10
20
30
40
50
60
70
2008 2010 2012 2014 2016 2018
Ca
pa
city
(Tb
/s)
15.5Tb/s155x100Gb/sHD-FEC
38.7Tb/s155x250Gb/s16QAMSpectrum shapingSD-FEC
65Tb/sProbabilistic shaped64QAMNonlinear mitigation
42 © Nokia 2016
• Highest capacity over submarine distance.
- Probabilisticconstellation shaping
- Nonlinear mitigation
- Adaptive rate FEC
<Change information classification in footer> © Nokia 2016
65Tb/s transmission over 6,600kmThanks to Probabilistic Constellation Shaping and NL mitigation
Highest capacity, close to expected limits
1
2
3
4
5
6
7
0 5 10 15 20 25
GM
I [b
its/
sym
bo
l/po
lar.
]
PS64QAM
SNR [dB]
Shannon
5.4
64QAM
Probabilistic constellation shaping: PCS 64QAM
43 © Nokia 2016
• Optical channel is not an arbitratry white Gaussian noise (AWGN) channel.
• Kerr nonlinearities because of propagation through optical fiber.
<Change information classification in footer>
Nonlinear mitigation
TX CDComp
MIMO2x2
Phase Est.Fiber optic
channel
062
.2
2
3
3
32
22 =+
∂∂−
∂∂−+
∂∂
AAT
Ai
T
AA
i
z
Ai γββα
Kerr nonlinearities
Attenuation Dispersion
FEC
44 © Nokia 2016
• Optical channel is not an arbitratry white Gaussian noise (AWGN) channel.
• Kerr nonlinearities because of propagation through optical fiber.
<Change information classification in footer>
Nonlinear mitigation
TX MIMO2x2
Phase Est.
FECFiber opticchannel
062
.2
2
3
3
32
22 =+
∂∂−
∂∂−+
∂∂
AAT
Ai
T
AA
i
z
Ai γββα
Kerr nonlinearities
Attenuation Dispersion
CD
NL
CD
NL
CD
NL
45 © Nokia 2016
• Direct detection
• Basics of coherent detection
• Implementation of coherent transponder
• System performances obtained thanks to coherent
• Conclusion
OUTLINE
46 © Nokia 2016
• Coherent detection has revolutionized the world of optical communication.
- Removed the need for optical dispersion compensation
- Solved the PMD issue
- Practical way to use the 2 polarization of the light
- Drastically Increased the spectral efficiency
• We are now approaching the spectral efficiency limits of fiber optics…
<Change information classification in footer>
Conclusion
400Gb/s on a single wavelength !