Abstract Terabit Optical Ethernet (TOE) will be affected

not only by limited bandwidth of information-infrastructure, but

also by its energy consumption. In order to solve both problems
simultaneously, in this invited paper, we describe several energy-
efficient (EE) hybrid-coded-modulation (CM) schemes enabling
TOE: EE-hybrid-multidimensional-CM, EE-generalized-OFDM,
and EE-spatial-domain-based-CM.
I. INTRODUCTION
The exponential internet traffic growth projections require
considerable increase of transmission data rates at every level
of the underlying information infrastructure, from core
networks to access networks. Higher volumes of traffic also
increase the energy consumption of transmission and
switching equipment needed to route this traffic. Recent
studies indicate that the energy consumed by the Internet
equipment is roughly 8% of the total energy consumed in the
US with predictions that it can grow up to 50%, with current
trend, by the end of this decade [1],[2]. Therefore, the Internet
is becoming constrained not only by capacity, but also by its
energy consumption.
In order to solve high-bandwidth demands and energy-
efficiency problems simultaneously, in this invited paper, we
describe several energy-efficient (EE) hybrid coded-
modulation (CM) schemes enabling Terabit Optical Ethernet:
energy-efficient multidimensional coded-modulation [3], EE
generalized-OFDM [4], EE 4D-multiband-coded-OFDM [5],
and EE spatial-domain-based-coded-modulation [6]. Due to
space limitations, the first two schemes are described here, the
other schemes will be presented at the conference. Common
properties of these schemes are: the employment of energy-
efficient coded-modulations, the employment of multiple
degrees of freedom for the conveyance of the information on
single-carrier optical signal and rate-adaptive coding based on
quasi-cyclic LDPC codes. These EE schemes are called hybrid
as they employ all available degrees of freedom for the
transmission over optical fiber including amplitude, phase,
polarization and orbital angular momentum (OAM). Namely,
from Shannons theory we know that the channel capacity is a
linear function in number of dimensions, and by increasing
the number of dimensionsbasis functions originating from
different degrees of freedom that are mutually orthogonal in
naturewe can dramatically improve the overall channel
capacity. On the other hand, the energy-efficiency problem
can be solved by properly designing the multi-dimensional
signal constellations such that the trans-information is
maximized, while taking the energy constraint into account.
These hybrid coded-modulation schemes are very flexible, as
they can be used for various applications ranging from short-
haul to long-haul, and can be used in SMF, MMF and free-
space optical links.

This paper was supported in part by NSF under grant CCF-0952711, and in
part by NEC Labs.
II. ENERGY-EFFICIENT SIGNAL CONSTELLATION DESIGN
The basic EE optical communication problem can be
formulated as follows. The symbols to be transmitted over
optical channel of interest are chosen from the followings set
X={x
1
, x
2
,,x
M
}. The symbols from this set have energies
E
1
,,E
M
and occur with a priori probabilities p
1
,,p
M
[p
i
=Pr(x
i
)]. The symbols from the set X satisfy the probability
constraint,
i
p
i
=1, and energy constraint,
i
p
i
E
i
sE. In the
presence of various channel impairments; such as fiber
nonlinearities, PMD, and PDL; and ASE noise; the
Lagranagian method can be used in maximizing the trans-
information (also known as mutual information) I(X,Y),
defined as I(X,Y)=H(X)-H(X|Y), where H(X) is the entropy of
the channel input X and H(X|Y) is the conditional entropy of
channel input X given the channel output Y. To determine the
optimum signal constellation, the energy-efficient Arimoto-
Blahut algorithm (EE-ABA) can be used as we described in
[3]. As an illustration, in Fig. 1 we report the information
capacities for different normalized energy cost functions for
number of amplitude levels per dimension L. In Fig. 1, we
provide the results corresponding to coherent detection. It is
clear from the Figure, that when the normalized energy cost
function is lower than one, we are facing certain information
capacity degradation.
8 12 16 20 24 28
1
2
3
4

C
a
p
a
c
i
t
y
,

C

[
(
b
i
t
s
/
c
h
a
n
e
l

u
s
e
)
/
d
i
m
e
n
s
i
o
n
]
Signal-to-noise ratio per symbol, SNR
s
[dB]
L=4:
normalized energy cost:
1 0.75
0.5
L=8:
normalized energy cost:
1 0.75
0.5
L=16:
normalized energy cost:
1 0.75
0.5

Fig. 1. Information capacities per dimension for different normalized
energy cost values and different number of amplitude levels per
dimension L in coherent detection case.
III. EE HYBRID MULTIDIMENSIONAL CM SCHEMES
The coordinates of the EE signal constellation from D-
dimensional mapper, implemented as a single look-up-table
(LUT), are used as the inputs to the D-dimensional modulator.
This modulator generates the signal constellation points by [3]

,
=1
= C ,
D
i D i d d
d
s f

(1)
Coding and Modulation Techniques Enabling
Multi-Tb/s Optical Ethernet
Ivan B. Djordjevic
University of Arizona, Department of Electrical and Computer Engineering, Tucson, AZ 85721, USA; E-mail: ivan@email.arizona.edu
171
MW1 (Invited)
1:30 PM 2:00 PM
978-1-4244-8939-8/11/$26.00 2011 IEEE

where |
i,d
denotes the dth coordinate (d=1,,D) of the ith
signal-constellation point, the set {u
1
,,u
D
} represents the
set of D=2MN orthogonal bases functions, where factor two
comes from two orthogonal polarization states, N denotes the
number of orthogonal OAM eigenstates and M basis
functions are defined as
u
m
(nT)=exp[j2t(m-1)nT/T
s
] (m=1,,M) (2)
where T
s
is the symbol duration, and T is the sampling
interval, related to symbol duration by T=T
s
/U, with U being
the oversampling factor. (In Eq. (1), C
D
denotes the
normalization factor.) The EE signal constellation coordinates
are split into N groups of 2M-coordinates per each OAM
mode. The 2M-coordinates of each group are used as input of
2M-dimesnional modulator, composed of two polarization-
multiplexed M-dimensional modulators. (Alternatively, one
2M-dimensional modulator can be used.) The M signal-
constellation point coordinates after up-sampling are passed
through corresponding discrete-time (DT) pulse-shaping
filters of impulse responses h
m
(n)=u
m
(nT), whose outputs are
combined together into a single complex data stream. After
separation of real and imaginary parts and digital-to-analog
conversion (DAC), the corresponding Re- and Im-parts are
used as inputs to I/Q modulator.
On receiver side, we first perform OAM mode-
demultiplexing, with OAM-demux block outputs representing
the projections along N OAM states. The nth OAM projection,
which is a 2M-dimensional signal, is used as input to the
polarization-beam splitter (PBS). The x-polarization (y-
polarization) signal, being itself an M-dimensional signal, is
used as input to the balanced coherent detector. After coherent
detection we recover Re- and Im-parts, which are after analog-
to-digital conversion (ADC) combined into a single complex
data stream. The same complex data stream is applied to the
inputs of M matched filters of impulse responses h
m
(n)= u
m
(-
nT). The corresponding outputs after re-sampling represent
projections along basis functions u
m
. At this point all 2MN-
coordinates of EE D-dimensional signal constellations are
estimated, and corresponding coordinate estimates
(representing the projections along D-basis functions) are
forwarded to the D-dimensional a posteriori probability
(APP) demapper, which calculates symbol log-likelihood
ratios (LLRs). From symbol LLRs we calculate the bit
likelihoods needed for LDPC decoding. For more details
about this scheme an interested reader is referred to [3].
The spectral efficiency of this hybrid D-dimensional
scheme, where D=2MN, is
-dim. constellation
2 2
PDM-QAM
2 QAM 2 QAM
log log
2log 2log
D D
E
E
S L D L
S M M
= =
(3)
times better than that of polarization-division-multiplexed
(PDM) QAM scheme. In (3), with M
QAM
we denoted the
QAM signal constellation size. Therefore, for the same
number of amplitude levels per dimension (M
QAM
=L
2
), the
spectral efficiency of the proposed scheme is (D/4)-times
better than that of PDM-QAM. The aggregate data rate (per
single wavelength) is determined by

2
2 s
ch. bits ch. sym. info. bits
log ( ) ,
ch. sym. s ch. bits
MN
L R r (4)
where r is the code rate, which is assumed to be equal for
LDPC codes at each level, and R
s
is the symbol rate. As an
illustrative example, by setting L=4, M=8, R
s
=50 GS/s, and
r=0.8, the aggregate data rate of 1.28 Tb/s per single OAM
mode is obtained. By varying the number of OAM modes up
to ten, the total aggregate data rate can be increased up to
fantastic 12.8 Tb/s per single wavelength. The numerical
results will be presented at the conference.
IV. EE GENERALIZED OFDM (EE-GOFDM)
The key idea behind the coded generalized OFDM (GOFDM)
[4] scheme is to increase the number dimensions over which
the signal constellation is defined in order to realize optical
transmission systems that are not only spectrally more
efficient but also more immune to channel impairments. We
call this scheme as a generalization of OFDM since we
consider the N orthogonal subcarriers of OFDM as a set of N
basis functions. The N-dimensional signal constellation is
obtained by using the concept outlined in Sec. II. We depicted
in Figs. 2(top) and (bottom) the transmitter and receiver
configurations for EE-GOFDM scheme. As it can be seen in
Fig. 2, when operated at the symbol rate of R
s
the GOFDM
scheme provides 2R
s
log
2
(L
N
) bits per second aggregate bit
rate, where the coefficient of two comes from the polarization-
multiplexing. As an illustrative example, when the symbol rate
is set to R
s
=25 GS/s (50 GS/s) and N=10, L=4 are used, the
aggregate bit rate reaches 1 Tb/s (2 Tb/s).
IFFT
Cyclic extension
insertion and
P/S conversion
LPF
LPF
DAC
DAC
Source
channels
(x-pol.)
1
b
Interleaver
bxn
b

N-dim.
symbols LDPC encoder
r=k / n
LDPC encoder
r=k / n
to fiber
Ix
LDPC-coded EE-GOFDM

EE N-dim.
mapper

4Dmodulator
Qx
Iy Qy

1
b
.
.
.
LDPC
decoder 1

.
.
.
N-dim. APP
Demapper
and
Bit LLRs
Calculator
Extrinsic LLRs
LDPC
decoder b
From
SMF
Local
Laser
PBS
PBS
LPF
LPF
ADC
ADC
FFT

Balanced coherent
detector (x pol.)
Cyclic extension
removal and
and S/P conversion
Balanced coherent
detector (y pol.)

Fig. 2 EE-GOFDM architecture: (top) Tx and (bottom) Rx configurations.
The common denominator of EE multidimensional CM
schemes described above is the employment of rate-adaptive
quasi-cyclic (QC) LDPC codes in FEC, which can adapt FEC
strength based on channel conditions. The idea is to deliver
data with target BER regardless of the destination.
REFERENCES
[1] B. G. Bathula, M. Alresheedi, and J. M. H Elmirghani, Energy efficient
architectures for optical networks, in Proc. IEEE London
Communications Symposium, London, Sept. 2009.
[2] N. Vasic, and D. Kostic, Energy-aware traffic engineering EPFL
Technical Report, 2008.
[3] I. B. Djordjevic, Energy-efficient spatial-domain-based hybrid
multidimensional coded-modulations enabling multi-Tb/s optical
transport, Opt. Express, accepted for publication.
[4] I. B. Djordjevic, M. Arabaci, L. Xu, T. Wang, Generalized OFDM
(GOFDM) for ultra-high-speed optical transmission, Opt. Express, vol.
19, no. 7, pp. 6969-6979, 2011.
[5] I. Djordjevic, H. G. Batshon, L. Xu, T. Wang, Four-dimensional optical
multiband-OFDM for beyond 1.4 Tb/s serial optical transmission, Opt.
Express, vol. 19, no. 2, pp. 876-882, 2011.
[6] I. B. Djordjevic, M. Arabaci, L. Xu, T. Wang, Spatial-domain-based
multidimensional modulation for multi-Tb/s serial optical transmission,
Opt. Express, vol. 19, no. 7, pp. 6845-6857, 2011.
172