Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
WOCC 2014
1 STMircroelectronics
I. I NTRODUCTION
Optical Orthogonal Frequency-Division Multiplexing
(OOFDM) is considered as the most promising technology
for next generation optical systems due to its high spectral
efciency and its immunity to inter-symbol interferences (ISI).
For example, for next generation passive optical networks
(NG-PON2), OFDM access based PON (OFDMA-PON) has
been shown to provide the highest upstream and downstream
rates with the highest passive reach [1], [2]. For low cost
approach, intensity modulation direct detection (IM/DD)
systems are preferred for both wireless and wired OFDM
based optical links. In IM/DD systems, the electric signal
is used to directly drive the optical source, which implies
that the OFDM signal applied to the optical transmitter
must be real and positive. Various forms of unipolar OFDM
schemes compatible with IM/DD systems were proposed in
the literature. DC biased optical OFDM (DCO-OFDM) [3][5] and asymmetrically clipped optical OFDM (ACO-OFDM)
[4]- [7] are commonly used. The DCO-OFDM technique
consists on adding a DC bias to the real OFDM signal and
hard clipping the remaining negative values. This technique
is simple to implement but it suffers from optical power
inefciency due to the added DC bias. Moreover, the hard
clipping results in a signicant clipping noise, especially for
low bias levels (7dB) and large constellation sizes. An optical
O3.3.pdf
63
DQG
6HULDO
'DWD
+HUPLWLDQ
0DSSLQJ
Symmetry
WOCC 2014
[>@
;>@
,))7
;>1@
'
$
3
[>1@ 6
[W
&OLSSLQJ
2SWLFDO
&KDQQHO
7UDQVPLWWHU
3
6
6HULDO 'HFRGLQJ
'DWD
<>@
\>@
))7
\>1@
<>1@
6
3
$
'
5HFHLYHU
Fig. 1. The block diagram of an IM/DD system based on optical OFDM
k = 0, 1, ..., N2 1,
(1)
N
1
k=0
X(k)exp(j2
kn
),
N
n = 0, 1, ..., N 1. (2)
signal amplitude can be approximated by a Gaussian distribution. Therefore, in order to avoid an excessive DC bias and
minimize the required optical power, an approach is to use
a DC bias Kb proportional to the root mean square RMS of
x(t), .
Kb = k,
(3)
where k is the clipping factor and 2 = E x2 (t) is the
variance of x(t). This is dened in the literature [4] as a bias
of 10 log10 (k 2 + 1)dB. The DC-biased time signal, xDC (t),
is given by
xDC (t) = x(t) + Kb .
(4)
The remaining negative peaks are clipped at zero to ensure
non-negativity of the time signal at the optical transmitter input, thereby resulting in clipping noise. The obtained unipolar
signal can be expressed as,
xu,DCO (t) = x(t) + Kb + nc (Kb ),
(5)
Xk ,
k odd
0,
k even
(6)
O3.3.pdf
N
N
) = x(n), n = 0, 1, ...,
1.
(7)
2
2
The anti-symmetric signal is made positive by hard clipping
all the negative values. The continuous time unipolar signal
can be expressed as
x(t), if x(t) > 0
xu,ACO (t) =
(8)
0,
if x(t) 0
x(n +
Thanks to the anti-symmetric property, clipping the negative values does not cause any loss of information, and the
transmitted odd data, X(k), can be integrally recovered from
the clipped signal. In ACO-OFDM, only the positive part of
the anti-symmetric bipolar signal is transmitted, with no need
to add a DC bias. Hence, the ACO-OFDM technique has more
optical power efciency than DCO-OFDM. However, for N
IFFT points, only N/4 subcarriers are used to carry data due
to Hermitian symmetry and the odd subcarriers modulation
constraint. As a consequence, the ACO-OFDM technique is
50% spectrally less efcient than DCO-OFDM.
III. A SYMMETRICALLY COMPANDED DCO-OFDM
A. System model
The block diagram of the asymmetrically companded DCOOFDM is shown in Fig. 2. First, the DC bias Kb is computed
from the real discrete time signal at the IFFT output, x(n).
Kb = k E {x2 (n)}.
(9)
In conventional DCO-OFDM systems, the DC bias is directly
added to the continuous time signal at the D/A converter
output and a hard clipping is performed in order to set to
zero the negative amplitudes. However, the dynamic range
of the bipolar real ODFM signal increases with the spectral
efciency. Thus, if a moderate DC bias is used, more negative
peaks are clipped, resulting in a higher clipping noise penalty.
As a result, for large constellations ( 64-QAM), the DCOFDM technique becomes inefcient in terms of BER. To
overcome this inefciency, an approach is to increase the
added DC bias [4], so that less negative peaks remain after
the addition of the DC bias and the clipping noise penalty
is reduced. However, this approach results in a high required
optical power to transmit the unipolar signal and therefore in
higher optical power inefciency. In this paper, we propose a
new technique to minimize the clipping noise with a moderate
DC bias. This technique consists on applying an asymmetric
companding transform to the time signal at the IFFT output,
so that the negative part is compressed and less negative peaks
are clipped even if a moderate DC bias is used. The discrete
time companded signal is given by
x(n),
if x(n) >
xc (n) =
(10)
x(n), if x(n)
where [0, 1] and [min {x(n)} , 0]. The DC bias
computed from the IFFT output signal is next added to the
companded signal, as shown in (11).
xDC (n) = xc (n) + Kb ,
n = 0, 1, ..., N 1.
(11)
WOCC 2014
[>@
,))7
[F>@
&RPS
[>1@
[F>1@
$GG '&
&OLSSLQJ
3
6
'&
\H>@
([S
))7
\H>1@
'
$
7UDQVPLWWHU
\>@
6
'&
5HPRYH
\>1@ 3
$
'
5HFHLYHU
O3.3.pdf
[ Q
[Q
x (n) =
.
(15)
0,
otherwise
The positive and the negative parts can be approximated
by one-sided Gaussian distributions with means + and ,
respectively.
WOCC 2014
and the receiver or adaptive. When they are xed, the companding process is independent of the OFDM signal statistical
characteristics and the parameters are not optimized to have
better performances. In adaptive companding transforms, the
parameters are computed adaptively so that the companding
process performs optimal performances in terms of PAPR and
BER. However, this technique results in data rate degradation,
because the updated parameters need to be transmitted to the
receiver in order to update the inverse companding transform.
In the proposed companding function, the parameter is
equal to the mean of the negative part of the bipolar signal
(i.e. = ). According to (16) and (17), | | = |+ |.
Furthermore, since the positive part is not modied by the
companding process, the means of the positive part before
and after the companding transformation are the same. On the
other hand, in wired IM/DD systems, the intensity modulation
can be done using either a Distributed Feedback laser (DFB) or
a Vertical-Cavity Surface-Emitting Laser (VCSEL). Contrary
to DFBs, which have a linear transfer function, VCSLEs
exhibit a nonlinear behavior [14]. However, as shown in [15],
when a dynamic signal is used to modulate the VCSEL around
the DC bias current, the VCSEL characteristic is almost a
linear function. The electric to optical converter (DFB or
VCSEL) being the main source of nonlinearity, the clipping
of the positive peaks is negligible. As a result, the mean of the
positive part is affected neither by the companding transformation nor by the clipping distortion. Thus, since = and
= + , the parameter at the receiver can be directly
extracted from the received companded signal. Meanwhile, the
parameter is xed according to the used constellation in
order to perform optimal clipping noise reduction. As a result,
no additional transmission data is needed.
IV. S IMULATION RESULTS
A. Clipping noise reduction
In the proposed companding technique, the negative part
of the bipolar OFDM signal is compressed, so that after the
addition of the DC bias, the DC-biased companded signal
has less negative peaks. Fig. 4 shows the amplitudes of a
conventional DC-biased OFDM signal and a companded DCbiased OFDM signal. The added DC bias in the two cases is
equal to 5dB.
=
=
E[x (n)]
2
t
1
dt = ,
t exp
2 2
2 0
2
(16)
E[x (n)]
2
0
t
1
t exp
dt = .
2 2
2
2
7LPH
(a)
In the conventional companding transforms, the companding
function parameters are either once xed at the transmitter
'LVFUHWH
&RQWLQXRXV
(17)
'LVFUHWH
&RQWLQXRXV
$PSOLWXGH
$PSOLWXGH
7LPH
(b)
O3.3.pdf
&OLSSHG
,GHDO
,PDJ
,PDJ
&OLSSHG
,GHDO
5HDO
(a)
5HDO
(b)
&RQY4$0
3URS4$0
&RQY4$0
3URS4$0
&RQY4$0
3URS4$0
&RQY4$0
3URS4$0
%(5
WOCC 2014
(EHOHF1G%
In conventional DCO-OFDM systems, for low constellations (i.e. 4-QAM), the time OFDM signal has a small dynamic range which results in a negligible clipping noise. Since,
the dynamic range of the time OFDM signal increases with
the constellation size, the clipping noise becomes signicant
for 16-QAM and dominant for larger constellations. As a
consequence, the BER curves at for constellations larger than
16-QAM [4], [5]. In the proposed DCO-OFDM technique,
since the clipping noise is negligible for 4-QAM constellation,
the parameter is set to 1, so that the channel noise, which
is the dominant noise, is not amplied at the receiver. For
larger constellations, the companding transformation, with
< 1, reduces signicantly the remaining negative values
and therefore the clipping noise. Furthermore, the proposed
companding function was designed so that a small part of the
time signal is amplied at the receiver. Thus, the amplication
of the channel noise is less signicant compared to the clipping
noise cancellation. As a result, the BER curves do not at
for the constellations larger than 16-QAM and very low BER
levels can be reached using only moderate DC bias levels.
C. Optical power efciency
The BER as a function of Eb(elec) /N0 forms a theoretical basis to compare OFDM schemes. However, in optical
systems, the transmitted optical power is the main system
constraint. Thus, the optical energy-per-bit to noise power
spectral density metric, Eb(opt) /N0 , is usually used to evaluate the optical power efciency of optical systems [4], [7].
Contrary to the electric power, which is proportional to x2 (t),
the transmitted optical power is equal to the amplitude of the
transmitted signal. Thus, the average transmitted optical power
is given by
Popt = E {x} .
(18)
For a normalized optical power, the optical energy-per-bit
to noise power spectral density is dened as,
Eb(opt)
E 2 {x} Eb(elec)
=
,
N0
E {x2 }
N0
(19)
O3.3.pdf
where E x2 is the average electric power of the transmitted
unipolar signal. Fig. 7 compares the optical power efciency
of the proposed technique with conventional ACO-OFDM and
DCO-OFDM schemes.
&RQY$&24$0
&RQY'&24$0G%
3URS'&24$0G%
&RQY$&24$0
&RQY'&24$0G%
3URS'&24$0G%
%(5
(ERSW1G%
WOCC 2014