Professional Documents
Culture Documents
2013 Elastic Optical Networks The Global Evolution To
2013 Elastic Optical Networks The Global Evolution To
Bell Labs Technical Journal 18(3), 133–151 (2013) © 2013 Alcatel-Lucent. • DOI: 10.1002/bltj.21631
Panel 1. Abbreviations, Acronyms, and Terms
ASE—Amplified spontaneous emission OE—Optoelectronic
AWGN—Additive white Gaussian noise OFDM—Orthogonal frequency division
BER—Bit error ratio multiplexing
BPSK—Binary phase shift keying OSNR—Optical signal-to-noise ratio
CAGR—Compound annual growth rate OSPF—Open Shortest Path First
CAPEX—Capital expenditures OTN—Optical transport network
CELTIC—Cooperation for a Sustained European OXC—Optical cross-connect
Leadership in Telecommunications PCE—Path computation element
DAC—Digital-to-analog converter PCEP—Path Computation Element Protocol
DSP—Digital signal processing PDM—Polarization division multiplexing
DVFS—Dynamic frequency and voltage scaling PMD—Polarization mode dispersion
EO-Net—Elastic-Optical NETwork PSK—Phase shift keying
ERO—Explicit route object QAM—Quadrature amplitude modulation
FEC—Forward error correction QPSK—Quadrature phase shift keying
FPGA—Field programmable gate array RDWA—Routing, data rate and wavelength
GFP—Generic framing procedure assignment
GMPLS—Generalized Multiprotocol Label RSA—Routing and spectrum assignment
Switching RSVP—Resource Reservation Protocol
HDTV—High-definition television RWA—Routing and wavelength assignment
IA—Impairment-aware SA—Spectrum allocation
IETF—Internet Engineering Task Force SFP+—10 GbE small form factor pluggable
ILP—Integer linear programming SMF—Single-mode fiber
IP—Internet Protocol SP—Single polarization
ITU—International Telecommunication Union SSON—Spectrum switched optical network
ITU-T—ITU Telecommunication Standardization TE—Traffic Engineering
Sector VCO—Voltage controlled oscillator
LCoS—Liquid crystal on silicon WDM—Wavelength Division Multiplexing
MPLS—Multiprotocol Label Switching WSON—Wavelength switched optical network
ODU—Optical data unit WSS—Wavelength selective switch
spectral efficiency across the entire network. introduction of flexible functionality enables better
Tradeoffs between high capacity and reach are tar- utilization of deployed transport resources (within
gets of key interest. their physical limits) in responding to variable traffic
In addition, optical transport networks must demands, considering aspects of both bandwidth and
cope with an increasingly heterogeneous and hold time. More dynamic resource allocation permits
dynamic environment: 1) connection lengths that better accommodation of traffic fluctuations, limiting
vary from a few hundreds to several thousands of the degree of over-provisioning required to make the
kilometers, 2) bandwidth requests that vary from network resilient to failures and also easing network
tenths to tens of Gb/s, and 3) connection hold times upgrades. Such capabilities are also expected to mini-
that range from quasi-permanent to hour-long or mize energy consumption and cost.
even shorter timescales. It is well understood that The recent advent of coherent detection in
“static” optical networks are typically dimensioned Wavelength Division Multiplexing (WDM) networks
based on peak hour traffic with considerable over- and the use of digital signal processing (DSP) in
provisioning needed to ensure robust handling of receivers over the last few years has cleared the way
traffic demands for a number of years. Clearly, the for use of flexible and versatile transceivers in optical
TSP
transponders and routers → new resource allocation
B algorithms for (multi-layer)
OXC elastic networks
TSP
TSP
(2) Flexible transponder (TSP) A OXC C OXC
TSP
TSP
(4) Performance prediction
OXC OXC
(1) Optical cross-connect (OXC) E D → new models required
OXC—Optical cross-connect
TSP—Transponder
Figure 1.
Main challenges with introducing elastic properties in photonic networks.
Q1 π/2
FEC/mapper
28 Gbit/s
RF amp T.M.
Client
Laser
RF amp
28 Gbit/s I2
Q2 π/2
28 Gbit/s
RF amp
(b) Receiver
Local
oscillator
Carrier phase estimation
T.E
Chromatic dispersion
Carrier frequency
.
demultiplexing
Demapper/FEC
ADC sampling
compensation
Equalization
Polarization
T.M
estimation
-Polarization
Decision
°diverse 90
Client
.
hybrid
ADC—Anallog-to-digital converter
FEC—Forward error correction
PDM—Polarization division multiplexing
QPSK—Quadrature phase shift keying
RF—Radio frequency
T.E.—Traverse electric
T.M.—Traverse magnetic
Figure 2.
Block diagram of a 100 Gb/s PDM-QPSK transceiver.
a tunable clock reference. To obtain data rates of • Channel spacing adaptation. The challenge for the
25, 50, 75 and 100 Gb/s, the symbol rate will transceiver is rather limited to the need for a
vary respectively between 7, 14, 21 and 28 (fully) tunable laser and local oscillator. However,
Gbaud. Simulations have shown that the achiev- all filtering elements such as optical filters and
able optical reach is only weakly dependent wavelength selective switches must be tolerant
upon the chosen symbol rate [29]. Therefore, of a non-standard ITU grid. If such optical filters
one of the main interests in symbol rate adapta- are not good enough or if the chosen channel
tion is to trade-off capacity versus energy sav- spacing is very tight to reach high spectral effi-
ings. The next few sections will quantify the gain ciency, further constraints can be put on the
with day/night traffic fluctuations and will show transceiver. In the former case, signal processing
a prototype of energy-proportional transport. blocks can indeed help to compensate for physical
Polarization X
Polarization Y
Data rate @
1R 2R 3R 4R 6R
symbol rate R
Distance L @
3L 2L 1.5L L L/5
optimum power
Figure 3.
Modulation formats and achievable data rate and optical reach.
{∑ }
N Ν
get quality. Hence, by reducing the FEC effi- 1 = var
______
SNRNL
uk = ∑ var(uk)
ciency, power savings are possible and promising. k=1 k=1
+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+
m Reserved
+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+
(a)
+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+
Grid =
CS Identifier m
DWDM
+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+–+
(b)
CS—Channel spacing
DWDM—Dense Wavelength Division Multiplexing
GMPLS—Generalized Multiprotocol Label Switching
RSVP—Resource Reservation Protocol
TE—Traffic Engineering
Figure 4.
Example of possible RSVP-TE protocol enhancement for implementing elastic optical networks in GMPLS
controlled networks.
bit rate per lane. Note that each of the four lanes −6
transmits at a rate B = 10⋅max(Bi)/4, i.e., the effec-
−7
tive rate of the busiest client sets the overall output
rate, tenfold. For practical constraints, we assume a −8 8 Gb/s
perfect load balancing between clients, i.e., the same 16 Gb/s
−9
effective rate Bi is sent to the ten clients. Even though 28 Gb/s
the above aggregation process is automated, the −10
8 12 16 20
operation speed of all the components on the mux-
OSNR
ponder needs to scale with B. We use a voltage con-
FEC—Forward error correction
trolled oscillator (VCO) to generate a centralized OSNR—Optical signal-to-noise ratio
clock and distribute it all across the aggregation mod-
ule and the transceiver. The transceiver is a typical Figure 6.
PDM-QPSK with 7 percent FEC overhead (using Bit error ratio performance versus OSNR.
Reed-Solomon) [9]. Hardware limitations of our real
time setup limit the maximum symbol rate to the theoretical linear inverse proportionality between
7 Gbaud, thus allowing a maximum aggregated bit noise tolerance and symbol rate. We attributed this
rate of 28 Gb/s. The receiver side comprises the usual to two factors. First, the electrical bandwidth limita-
optical front end of a polarization-diversity coherent tions of the transmitters and receivers in our proto-
receiver, whereby the signal is mixed with the out- type induced a penalty when working at 28 Gb/s,
put of a local oscillator before being converted to the where a BER floor can clearly be seen around 3x10−6.
electrical domain through four balanced photodi- Second, our prototype did not incorporate a rate
odes. Each resulting electrical signal is then digitized adaptive anti-aliasing filtering function in the optical
through analog-to-digital converters, sampling at front end, which would be required for optimal miti-
twice the baudrate. The sampled data are processed gation of optical noise, since it was instead designed
in field programmable gate arrays, where polariza- for the highest symbol rate. For example, a differ-
tion demultiplexing is performed with 9 tap finite ence of only 0.6 dB ONSR in sensitivity can be seen
impulse response filters, arranged in a butterfly at 10−4 BER when moving from 8 Gb/s to 16 Gb/s,
structure and updated by the constant modulus algo- whereas 3 dB would be expected with an optimal
rithm. Carrier frequency and phase estimation are anti-aliasing filter. We also measured the end-to-end
performed using the Viterbi algorithm, and finally Ethernet frame loss ratio, as shown in Figure 7. The
the symbols are converted to bits through hard deci- results are in line with the expectation for the chosen
sion. All the aforementioned operations are per- FEC (i.e., Reed-Solomon).
formed in real time. The detailed experimental setup Second, the power consumption of the receiver
and performance results can be found in [41]. DSP was measured as a function of symbol rate and is
We first investigated the muxponder prototype’s reported in Figure 8. The measurement was per-
sensitivity to noise as a function of data rate. In formed by direct inspection of the voltage and current
Figure 6 we report the pre-FEC bit error ratio as a supplied to the receiver board. A linear relationship is
function of the optical signal-to-noise ratio (OSNR). demonstrated between power consumption and
While we found that tolerance to noise increases actual transported traffic. This is typical of the power
when the symbol rate decreases, we did not observe consumption of logical gates versus clock rate, when
65
DSP power