JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO.

24, DECEMBER 15, 2010

3519

Wavelet Packet Transform-Based OFDM for

Optical Communications
An Li, William Shieh, Member, IEEE, and Rodney S. Tucker, Fellow, IEEE

AbstractWavelet packet transform-based optical OFDM

(WPT-OFDM) systems can be used as an alternative approach
to the conventional optical OFDM systems. In this paper, we
investigate the performance and limitation of real WPT-OFDM
systems. Its double-sideband characteristic makes it sensitive to
spectrally non-symmetric dispersion such as polarization-mode
dispersion (PMD). Simulations of dual-polarization WPT-OFDM
transmission are carried out in the presence of PMD, and compared with conventional Fourier transform-based optical OFDM
(FT-OFDM). The results show that WPT-OFDM is very sensitive
to PMD, incurring 1-dB penalty at 511 ps DGD for 112-Gb/s
dual-polarization transmission. PAPR and nonlinearity performance of WPT-OFDM are also analyzed in this work. We show
that the Haar wavelet has 0.9-dB improvement in nonlinear
launch power limit compared with FT-OFDM (CP = 1 8)
in a conventional transmission dispersion configuration where
inline dispersion is fully compensated by a dispersion-compensation-fiber (DCF).
Index TermsFIR digital filters, optical fiber dispersion, optical
fiber polarization, orthogonal frequency-division multiplexing,
quadrature mirror filters, wavelet transforms.

I. INTRODUCTION
PTICAL orthogonal frequency division multiplexing
(OFDM) is a promising modulation format that applies
OFDM modulation techniques to the optical transmission
systems. It has recently attracted much interest from the optical communication community [1][3] because of its high
spectral efficiency and robustness against impairments such as
chromatic dispersion (CD) and polarization mode dispersion
(PMD). Together with coherent detection, coherent optical
OFDM (CO-OFDM) has the benefits of the high spectral
efficiency and high receiver sensitivity. In CO-OFDM systems,
by appropriately choosing the length of cyclic prefix (CP) and
inserting training symbols at the transmitter, both inline CD and
PMD can be fully compensated via digital signal processing
(DSP). CO-OFDM transmission at 1 Tb/s has already reported
[4], [5].
Conventional OFDM systems are based on Fourier transforms (FTs). In these OFDM systems, low rate subcarriers

Manuscript received May 17, 2010; revised August 22, 2010, September 15,
2010; accepted October 17, 2010. Date of publication October 25, 2010; date
of current version December 03, 2010.
The authors are with the Centre for Ultra-Broadband Information
Networks (CUBIN), Department of Electrical and Electronics Engineering, The University of Melbourne, Melbourne, VIC 3010, Australia (e-mail: a.li2@pgrad.unimelb.edu.au; w.shieh@ee.unimelb.edu.au;
r.tucker@ee.unimelb.edu.au).
Digital Object Identifier 10.1109/JLT.2010.2089673

are orthogonally transformed into time domain signals, and

the orthogonality between subcarriers can prevent inter-carrier
interference (ICI). It has been shown that OFDM modulation/demodulation can be efficiently implemented using inverse
discrete Fourier transform (IDFT)/discrete Fourier transform
(DFT) electronically. Wavelet transforms (WTs), or wavelet
packet transforms (WPTs) in particular [6][8], are relatively
new concepts in transmission systems by which a signal is
expanded in an orthogonal set called wavelets [6][17].
Similarly to Fourier transform, wavelet transforms can provide
orthogonality between OFDM subcarriers, however the basis
functions are wavelets instead of sinusoids. Unlike sinusoids
that are infinitely long in the time domain, wavelets have finite
length. Therefore, WTs have both frequency and time localization. The incentive to use WPTs rather than FTs in OFDM
is to provide better spectral roll-off and to remove the need
for CP [18]. In addition, wavelets can provide more freedom
in system design. It has been proposed that wavelet packet
transform-based optical OFDM (WPT-OFDM) be applied as
an alternative approach to conventional Fourier transform based
optical OFDM (FT-OFDM), with better performance in short
haul [18]. It is shown that WPT-OFDM can mitigate a CD of
3,380 ps/nm at 112 Gb/s rate, a remarkable result without the
need of CP.
While [18] is based on single-polarization systems, dual-polarization transmission has been recognized as a promising technique for the future 100 Gb/s or even 1 Tb/s Ethernet transport
[19][26]. It is of great importance to study the system performance of dual-polarization WPT-OFDM. In this paper, we
discuss the PMD impact and PAPR-induced nonlinearity performance of WPT-OFDM systems. We first introduce theoretically some fundamentals of WPT-OFDM and point out its potential deficiency when applied to the optical channel. Then we
assess the impact of PMD on WPT-OFDM transmission with
various commonly used wavelets, and comparison is drawn between WPT-OFDM and conventional FT-OFDM. We find that
WPT-OFDM is very sensitive to PMD, incurring 1-dB penalty
at 5 ps of differential-group-delay (DGD) as opposed to 76 ps
for conventional FT-OFDM without CP, both at a data rate of
112 Gb/s. We also point out that using complex wavelets may be
one solution to the PMD sensitivity problem of WPT-OFDM. It
is known that the nonlinearity performance of OFDM systems
is expected to be worse than single carrier systems due to its
high peak-to-average power ratio (PAPR) of transmitted timedomain signals. However for WPT-OFDM, different wavelets
may provide different PAPR due to the waveform shaping. We
find out that the Haar wavelet [13] the lowest probability of
high PAPR thus the best nonlinearity performance. Then we

0733-8724/$26.00 2010 IEEE

3520

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 24, DECEMBER 15, 2010

verify this conclusion through simulation of signal transmission

over a fiber link where inline dispersion is 100% or 95% compensated by DCF. The result shows that the nonlinear limit of
the Haar wavelet has 0.9 dB improvement over the FT-OFDM
.

. In this expanded subspace, if

for all
expressed as

II. WAVELET PACKET TRANSFORM FUNDAMENTALS

A. Fourier Transform (FT) and Wavelet Transform (WT)
It is instructive to start our discussion with the Fourier transform that is more well-known in optical communications than
the wavelet transform. The Fourier transform is an operation
that transforms one complex-valued function into another.
Since the original function is typically of time-domain and the
new function is of frequency-domain, the Fourier transform is
often called the frequency-domain representation of the original function. Discrete Fourier transform (DFT) is a specific
kind of Fourier transform that requires the input function be
discrete, which means its non-zero values have a finite duration.
Mathematically, forward and inverse DFTs are defined as
(1)

, it can be

(8)
For
the span
is larger then
since
has finer
scale and thus can represent finer detail, and vice versa for
. This change of scale can be considered as a change of resolution. Wavelet satisfies a so-called multiresolution formulation requirement [9], [11] which is designed to represent signals
where a single event is decomposed into finer and finer detail.
The basic requirement of multiresoultion analysis (MRA) has
the form given by
(9)
with
and
, which means a space containing higher resolution will also contain those of lower reso, then
. The nesting
lution. Therefore, if
(spans of
) can be achieved by requiring that
of
. This means that if
is in its also in (spans
). Namely,
can be expressed by a weighted sum of
of
as
time-shifted

(2)

(10)

It is known that DFT only has frequency localization, and its

basis functions, the sinusoids are infinitely long in time domain.
In comparison, the wavelet transform is the representation of
a function by an orthogonal set called wavelets. It is another
form of time-frequency representation for the original function.
In order to best understand wavelet and wavelet transform, we
start by defining the scaling function and its time shift set given
by [11]

where
is a sequence of real or complex numbers called the
scaling function coefficients(or scaling filter).
is an expanded space of , we define a new set of
Since
as the orthogonal complement of in
, that is
spaces

(3)

(12)

is the vector space

where is the set of all integers, and
is a space spanned by these
of square integrable function.
scaling functions that defined as
(4)
The overbar means closure, namely
(5)
By changing the time scale of the scaling functions, one can
increase the size of the subspace, for example, from to by
a two-dimensional parameterization (time and scale) of scaling
function
(6)
The expanded subspace

is spanned by the new function as

(7)

(11)
It follows from (11) that

which is shown in Fig. 1. It follows that

(13)
Now we will define the wavelet function
. Similarly to the
be a space spanned by a wavelet funcscaling space , let
tion
, and for the expanded space
spanned by
using expressions equivalent to (4)(7). Since
is orthogonal
and
is
to , orthogonality between scaling function
required, which means
(14)
Because these wavelets are in the space spanned by the next finer
, similar to the scaling function
scaling function, e.g.,
can also be expressed by a weighted sum of timeshifted
as
(15)

LI et al.: WAVELET PACKET TRANSFORM-BASED OFDM

3521

transform (DWT). The discrete wavelet expansion of any signal

is given by

(23)

Fig. 1. Scaling function and wavelet vector spaces.

where
is called the wavelet function coefficients (or
wavelet filter). The relationship between these two coefficients
are restricted by the orthogonality condition (11), given by
(16)
If

. is the set of all integers,

is the vector
for
is an arbitrary inspace of square integrable function, and
teger. It can be seen that and provide the frequency (or scale)
also known as detail coefficient, and
and time localization.
also known as approximation coefficient in the wavelet
), which can be obtained
expansion (forward DWT of signal
from the following inner products
(24)

has a finite even length

(17)

In addition, for the wavelet coefficients

, an orthonormal
perfect reconstruction (PR) requirement usually applies. PR
means that in the absence of encoding, quantization and
transmission errors, the reconstructed signal can be perfectly
reassembled at the receiver, which is critical for transmission
system. The orthonormality condition and the perfect reconstruction condition of the orthonormal two bands PR filter are
[9]
(18)

(19)
where in (18) is the vector form of
.
We now use Haar wavelet to explain the wavelet and scaling
is
coefficients. Haar wavelet function
(20)
and its scaling function is
(21)
The filter coefficients of Haar wavelet can be obtained by applying (10) and (15) given by
(22)
Now we have defined scaling and wavelet function as well
as scaling and wavelet coefficients. From the wavelet theory we
know that for any arbitrary signal we can expand it into a sum of
scaling and wavelet functions, and this process is called wavelet
transform (WT). Similarly to the Fourier transform, wavelet
transform also has a discrete analogy called discrete wavelet

(25)

Combing (10) and (24), (15) and (25), following relationship

can be deduced [11]
(26)
(27)
can be efUsing (26) and (27), the DWT of a signal
ficiently computed using discrete-time filterbanks that are either infinite-time response (IIR) or finite time response (FIR)
filters and [8][14]. In practical applications and for computational efficiency, one prefers a wavelet with compact support
and wavelet function
can
where the scaling function
be considered finite in length. Detailed filter bank implementation of the DWT algorithm is as follows: To start the DWT,
at high resoluone needs to get the detail coefficients
acts
tion and for high enough scale the scaling function,
as delta function with the inner product as a sampling of
according to (24). Therefore, the samples of
are passed
through a low-pass filter (scaling filter) and high-pass filter
(wavelet filter) simultaneously, resulting in a convolution of
the two. The two filters are related to each other and they are
known as a quadrature mirror filter (QMF); the filter outputs are
then down-sampled by 2 since half the frequencies of the signal
have been removed, half of the samples can be discarded according to Nyquists theory; the outputs will give the detail co(from the high-pass filter ) and approximation
efficients
(from the low-pass filter ) as we show from
coefficients
(26) and (27); this decomposition process can be repeated to
further increase the frequency resolution, but only the approximation coefficients are decomposed. The above implementation of algorithm can be represented as a lower-half binary tree
structure as shown in Fig. 2. It is important to notice that for a
-point DFT, the bandwidth is uniformly divided; however for

3522

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 24, DECEMBER 15, 2010

Wavelet packet functions

the recursive relationships

for

are defined via

(30)

Fig. 2. Block diagram of a discrete wavelet transform (DWT) with 3-level filter
banks. # 2 stands for two times downsampling. f(t ) at the input is the sampled
input signal f(t).

(31)

We show in part A that

sible orthonomal basis for
tions

. Another posis the set of wavelet packet func-

(32)
If we also require
as the orthogonal complement of
in
, as in (11), that is
, it can be seen
is
that an orthonomal basis for the complementary space
(33)

Fig. 3. (a) Fourier transforms with uniform division of bandwidth. (b) Wavelet
transforms with logarithmic division of bandwidth.

an n-level DWT, the bandwidth is logarithmically divided since

only half of the spectrumthe low pass filter outputs are decomposed at each level. An explicit comparison of the bandwidth
division feature between DFT and DWT is shown in Fig. 3.
B. Wavelet Packet Transform (WPT)
The logarithmic division of the bandwidth in the wavelet
transform is not well suited for multicarrier communication
such as OFDM systems [15]. Wavelet packet transforms are
a generalization of wavelet transforms where the orthogonal
basis functions are wavelet packets [7], [8]. As we have
previously discussed, in DWT process, each level is calculated
by passing only the previous approximation coefficients
through high and low pass filters. A discrete wavelet packet
transform (DWPT), also called a wavelet packet decomposition
(WPD), on the other hand, decompose both the detail and
approximation coefficients at each level. Therefore, DWPTs
have more flexibility in tree structure where the bandwidth can
be arbitrarily (or uniformly, which is more commonly used)
divided according to tree pruning [7], [8]. In order to explain
the concept DWPT, first we define a set of wavelet packet
as [17]
functions

The orthonormal bases

and
in DWT we already familiar with can be re-written in terms of
for and
this new notion as
for .
There are many other bases that can be used resulted from
the appropriate selection of combination of the indexes of
. Precisely speaking, a basis for
can be
formed by allowing to range over , and choosing an index
set
so that the
are disjoint and cover the entire
intervals
, which means
interval
(34)
This equation can be thought as equivalent to covering the entire
time-frequency plane with windows of various shapes. It can be
easily seen that wavelet basis forms such a cover. However, for
wavelet packet, the collection of all wavelet packet functions
contains too many elements to
form an orthonormal basis. We need to carefully choose a subset
of this collection in order to obtain a proper basis. Assuming
there is a suitably chosen set of indexes denote by , we can
into its wavelet packet
decompose any signal
components by [17]

(28)
(no subscripts) is to have
. The extra index
is called the modulation parameter or oscillation
parameter. The first two wavelet packet functions are known as
the usual scaling function and wavelet function

(35)

where

the coefficients can be computed via

(36)

(29)

LI et al.: WAVELET PACKET TRANSFORM-BASED OFDM

3523

Fig. 4. Implementation of discrete wavelet packet transform (DWPT). # 2 stands for 2 times down-sampling.

Fig. 5. Implementation of inverse discrete wavelet packet transform (IDWPT). " 2 stands for 2 times up-sampling.

Combining with (26) and (27), we can obtain the following relationship
(37)
(38)
and this is equivalent to a -level full wavelet packet decomposition (full binary tree structure).
For OFDM systems that require uniformly division of bandwidth, a WPT with a full binary tree structure is utilized (see
Fig. 4). Detailed implementation of the DWPT algorithm is
given as follows: the n-level DWPT process has a binary tree
high and low FIR filters ( and
structure consisting of
) at level m. Similar to DWT, the root furthest to the left is
the sampled time-domain signal. It is first split into two equal
sequences, then convoluted with the decomposition high-pass
filter (or low pass filter ), followed by 2 times down-sampling. The high- and low-pass branches are the new inputs for

the next level. After n levels of such iterative processes, the

leaves furthest to the right are the decomposed wavelet packet
coefficients.
The inverse discrete wavelet packet transform (IDWPT), also
called the wavelet packet reconstruction (WPR), has a mirror
image process of the DWPT with a similar tree structure, where
the dataflow are from leaves to the root, as shown in Fig. 5.
The leaves furthest to the left are the packet coefficients, followed by convolution with reconstruction high-pass filter
(or low-pass filter ). The high- and low-pass branches are
then summed up generating a new sequence. After n levels of
such iterative processes, the root furthest to the right gives the
time-domain transformed data. For data transmission, DWPT
and IDWPT must be used as a pair with the reconstruction and
decomposition filters having the following relationship [9]
(39)
(40)

3524

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 24, DECEMBER 15, 2010

Fig. 6. Conceptual diagram of WPT(FT)-OFDM system setup. For WPT-OFDM, IDWPT and DWPT are used as a pair, and for FT-OFDM, IDFT and DFT are
used as a pair. PBC/PBS: Polarization Beam Combiner/Splitter; LPF: Low Pass Filter; GI: Guard Interval. Cyclic prefix can be inserted for FT-OFDM but not
available for WPT-OFDM.

III. PRINCIPLE OF OPTICAL WPT-OFDM

In order to best understand the principle of optical
WPT-OFDM, we first examine the principle of conventional
FT-OFDM. In a FT-OFDM system with coherent detection
(CO-OFDM) [1], at the transmitter, block based sampled
data are mapped onto the complex domain through QPSK
or higher-order QAM modulation with each channel corresponding to an orthogonal frequency domain subcarrier and
transformed into time domain signal though IDFT, then up-converted to optical domain using an optical IQ modulator for
fiber transmission [26]. At the receiver, the coherent received
optical signal is down-converted to the electrical domain, then
sampled and transformed back to frequency domain through
DFT for equalization and decision. As we have previously
discussed, WPTs have similar time-frequency representations
that can also provide orthogonality between OFDM subcarriers
as FTs, therefore, a simple idea is to use DWPT to replace DFT
in the system setup to obtain WPT-OFDM. A conceptual block
diagram of WPT-OFDM is given in Fig. 6. The mechanism
of WPT-OFDM in optical transmission is as follows: Similar
to FT-OFDM where one channel occupies a unit frequency
(subcarrier), in WPT-OFDM, each channel occupies a wavelet
packet (subcarrier in wavelet domain). IDWPT are used at
the transmitter that reconstructs the time domain signal from
wavelet packets. DWPT are used at the receiver to decompose the time domain signal into different wavelet packets by
means of successive low-pass and high-pass filtering in the
time domain as described in the last section. The prominent
difference between IDWPT and IDFT is that IDWPT is not
block based, namely, in (I)DWPT, the input and output data
are continuously flowing in and out, and there is no need to
delay to store one complete block to process as in (I)DFT. For
FT-OFDM, sinusoids are infinitely long in the time domain, so
a windowing has to be applied, which will lead to out-of-band
radiation and performance degradation because of inter-symbol
interference (ISI). An efficient technique to mitigate this degradation is to introduce a cyclic prefix (CP)a duplicate of later
part of the time-domain OFDM signal which is appended to

the beginning of each OFDM block. Properly designed CP significantly increases the tolerance to or completely removes the
CD impairment. However for WPT-OFDM, its basis function
wavelets are finite in time. The inter-symbol orthogonality
in wavelet transform is maintained due to the shift orthogonal
property of the waveforms. In the wavelet transform, symbols
are overlapped in time domain. This feature increases the
symbol duration which provide CD tolerance, but also prevents
the availability of CP [18]. A comparison of computational
complexity between WPT- and FT-OFDM is dependent on the
specific family of wavelet and system configuration, and has
been discussed in detail in [18].
IV. SYSTEM MODEL IN THE PRESENCE OF PMD
Wavelet transforms are generally defined in the real-value domain but can be also defined in the complex domain, solely depending of the scaling and dilatation filter coefficients. Realvalued wavelets are predominantly designed to process real signals such as patterns and images [13]. However for a fiber optic
channel, whether the input signal is modulated by real or complex signal, the up-conversion to the optical domain will inevitably generate two spectral sidebandsone positive and the
other negative. Fig. 7 shows the spectrum of Daubechies 32
(db32) wavelet [13], a typical wavelet with positive and negative
sidebands. In order to evaluate the PMD impact on WPT-OFDM
and compare with FT-OFDM counterpart, we use the model
of CO-OFDM transmission in a 2 2 multiple-input multipleoutput (MIMO) representation [25]. In optical WPT-OFDM, the
received sampled time-domain signal is transformed to wavelet
domain through DWPT (Fig. 6), and the received wavelet packet
coefficients (OFDM information symbol) in a form of Jones
vector for the th wavelet packet (subcarrier) in the th OFDM
symbol can be written as [25]
(41)
is the transmitted symbol as a Jones vector and
is the corresponding received noise vector.
is the
OFDM symbol dependent phase noise. Phase dispersion due to
fiber chromatic dispersion (CD) is given by
where

LI et al.: WAVELET PACKET TRANSFORM-BASED OFDM

3525

Fig. 7. Spectrum of one db32 wavelet packet, constructed using 5-level

IDWPT.

Fig. 8. Illustration of PMD impact on k th wavelet packet in WPT-OFDM systems for x polarization component.

(42)
For simplicity, we use the commonly-used first-order PMD approach modeled as
(43)
(44)
where is the DGD of the link, and are the polar and azimuth angle of the principle state of polarization (PSP) respecis the subcarrier frequency. In FT-OFDM systively, and
tems, the individual subcarrier, as the orthogonal basis of the
in (42)
Fourier transform is single-sideband by nature,
in (43) can be conveniently estimated and compenand
sated. In contrast, in WPT-OFDM systems, the modulated signals are double-sideband by nature (see Fig. 7). By applying
theory similar to (41)(44), chromatic dispersion influence on
WPT-OFDM is benign because the two sidebands experience
. However PMD
equal phase dispersion
does not hold such phase symmetry. The Jones matrixes for
the positive and negative sidebands do not equal, i.e.,
. Upon reception where the two sidebands need to be recombined and projected onto real wavelet basis, the two sidebands experience two different dispersions, and the addition
of the two does not reproduce the real-wavelet basis, resulting
in violation of the orthogonality and therefore inter-packet-interference. This is illustrated in Fig. 8 where the conceptual
figure of PMD impact on OFDM subcarriers with x-polarization launch is shown. We conclude that WPT-OFDM will be
more susceptible to PMD than conventional FT-OFDM.
V. SIMULATED SYSTEM DESCRIPTION
We have carried out numerical simulation to compare the
transmission performance of dual-polarization WPT-OFDM

Fig. 9. Baseband spectrum of simulated 112 Gb/s dual-polarization

WPT-OFDM system using db32 wavelet.

systems with FT-OFDM systems. The simulation parameters

are as follows: Dual-polarization OFDM data rate at 112 Gb/s,
64 subcarriers fully filled with data occupies a 28 GHz bandwidth in spectrum and oversampled by a factor of 2 to avoid
aliasing. Different wavelets from Haar, Daubechies, Coiflet and
Johnston family [13], [14] are chosen for wavelet transform
and compared with Fourier transform. Fig. 6 shows the signal
flow of a typical WPT-OFDM system used in simulation. At
the transmitter, the serial PRBS at 56 Gb/s is converted into
64 parallel data pipes, mapped onto the complex plane in
QPSK modulation, each corresponding to the wavelet packet
coefficient in frequency domain. The wavelet coefficients
are converted into serial time-domain wavelet signal through
IDWPT. The wavelet signal is then up-converted onto an
optical carrier with central frequency at 193.1 THz using an
ideal optical I/Q modulator. The two 56-Gb/s optical signals
are polarization combined into a 112-Gb/s WPT-OFDM signal
and launched into the optical fiber.
Fig. 9 shows the baseband spectrum of the WPT-OFDM
signal with a bandwidth of 28 GHz. The WPT-OFDM signal
is then passed through a fiber with chromatic dispersion and
PMD. At the receiver, the WPT-OFDM signal is coherently
down-converted to the RF domain, sampled and transformed
back from time to frequency domain through the DWPT,
followed by the channel equalization, symbol decision, and
bit-error-ratio (BER) estimation. 10 training symbols with
alternative polarization launch are sent for channel estimation.
A one-tap equalizer is introduced for equalization of CD and
PMD at the receiver [25]. In all simulation and analysis for
the PMD impact, we have assumed that the signal on each
polarization is launched at 45 with respect to the PSP of the
PMD for which we find the worst penalty takes place.
VI. RESULTS AND DISCUSSION
Fig. 10 shows the simulated BER of WPT-OFDM using
different wavelets as well as FT-OFDM. For FEC limit at BER
and optical signal-to-noise ratio (OSNR) of 20 dB,
of
Johnston64 (E) wavelet has the longest reach, indicating a remarkable over 5,600 ps/nm CD tolerance ( 330 km for SSMF
with
ps/nm/km). The naming convention of the
wavelet is its family name followed by the order. Fig. 11 shows
the BER performance versus OSNR at 0 and 10 ps of DGD
with different wavelets for a 112-Gb/s WPT-OFDM signal.
The required OSNR at FEC threshold as a function of DGD
for the same set of wavelets are analyzed and the results are
shown in Fig. 12. For 1-dB OSNR penalty, the DGD tolerance
is about 11, 6, 6, and 5 ps for Haar, Coifilet5, Daubechies32,
and Johnston64(E) wavelets, respectively. The performance of

3526

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 24, DECEMBER 15, 2010

Fig. 10. BER versus transmission distance for WPT-OFDM and FT-OFDM.
CD = 17 ps/nm/km. Wavelets are Haar, Coiflet, Daubechies [13] and Johnston
[14] family. Number of subcarriers Nsc = 64.

Fig. 12. Required OSNR at

Nsc = 64.

Fig. 11. BER versus OSNR for WPT-OFDM without DGD (0 ps) and with
DGD (10 ps). Number of subcarriers Nsc = 64.

Fig. 13. OSNR penalty versus DGD for FT-OFDM without CP and with CP
of 1/8.

OSNR penalty versus DGD for FT-OFDM systems is shown in

Fig. 13, indicating 132 and 80 ps DGD can be tolerated with
and without CP respectively. Consequently, the PMD tolerance
of FT-OFDM is more than seven times higher than that of
WPT-OFDM. By using sufficiently long CP, the PMD penalty
can be greatly alleviated or even eliminated in FT-OFDM
systems [25]. We also perform the simulation to study the
PMD tolerance dependence on the number of subcarriers for
WPT-OFDM and result is presented in Fig. 14. It can be seen
that for WPT-OFDM systems, almost no improvement can be
gained by increasing the number of subcarriers. This is because that the double-sideband characteristics of WPT-OFDM
spectrum will always adversely affect the PMD performance,
irrespective of how finely each wavelet is being partitioned.
In OFDM systems, nonlinearity performance is correlated
to the peak-to-average power ratio (PAPR) of the transmitted
RF domain OFDM signal. The complementary cumulative distribution function (CCDF) distributions of PAPR are plotted
in Figs. 15 and 16, for the non-oversampled and oversampled
scheme. The oversampling factor No is chosen to be 1 or 8.
We find out that without oversampling, the differences between
WPT-OFDM and FT-OFDM are minimal. Oversampling is usually applied in real OFDM system to avoid aliasing noise. When
8-times oversampling is applied we see that Haar wavelet has
the lowest PAPR. The nonlinearity performance is analyzed by
the BER at the receiver versus launch power into the fiber, as
shown in Figs. 17 and 18. In this simulation we only focus on
the relationship between PAPR and nonlinearity performance,

BER = 1e3 versus DGD for WPT-OFDM.

Fig. 14. OSNR penalty versus DGD for WPT-OFDM using Johnston wavelet
with Nsc of 64, 128, and 256.

PMD impact on nonlinear transmission performance is ignored

(for PMD impact we are more concern about the linear compensation).
For a fair comparison we use two different dispersion configurations, one is 100% dispersion compensated and the other
95% compensated. The optical link consists of two spans. In
ps/nm/km
each span we have one 100 km SSMF with
followed by a 14.94 km (100% comp.) or 14.2 km (95% comp.)
ps/nm/km. Power loss due to
DCF with
SSMF and DCF are fully compensated by two EDFAs with
noise figure at 6 dB. The launch power into SMF is swept from 4
dBm to 10 dBm to show the performance and nonlinearity limit.

LI et al.: WAVELET PACKET TRANSFORM-BASED OFDM

Fig. 15. CCDF of PAPR for WPT-OFDM and FT-OFDM.

sampling factor No = 1.

3527

Nsc = 64, over-

Fig. 16. CCDF of PAPR for WPT-OFDM and FT-OFDM. Nsc

8.

= 64 No =
;

Fig. 18. BER versus launch power after 2

span is 95% compensated by DCF.

2 100 km transmission. CD in each

plication is limited due to very low CD tolerance (see Fig. 10).

For wavelets with high CD tolerance such as Johnston wavelet,
PMD sensitivity becomes a major drawback. In order to solve
this problem of PMD sensitivity, one natural way is to generate single-sideband wavelets in frequency domain. Similar to
the Fourier transform that is based on complex-valued oscillating sinusoids, specifically designed complex wavelets with
complex-valued scaling function and wavelet function
can also have the same characteristics if
and
form a Hilbert transform pair [16]. This simple solution leads us to a path searching for a complex wavelet suitable for use for optical fiber channel. Unfortunately, complex
wavelet by itself is a relatively new field [16]. The analytical
solution of the complex waveletwhich needs to be realizable
by FIR filters from a practical point of viewis rarely found, or
perhaps impossible. Dual-tree structure can give some kind of
approximately single-side band characteristics, but will lead to
additional redundancy [16]. So the solution in terms of Hilbert
pair complex wavelet, or approximate Hilbert pair, is still an active research topic in the field of mathematics, and its adaptation
into optical communications remains an open question, which
we will explore in our future work.
VII. CONCLUSION

Fig. 17. BER versus launch power after 2 100 km transmission. CD in each
span is fully compensated by DCF. Nsc = 64; No = 8.

We find out that Haar wavelet has the best nonlinearity performance in the full dispersion compensation setup. The nonlinear
has been increased
limit of launch power for a BER of
, and 0.9 dB
by 1.7 dB compared with FT-OFDM
.
Although the Haar wavelet has the lowest sensitivity to PMD
and the best nonlinearity tolerance among all wavelets, its ap-

A wavelet packet transform-based optical OFDM

(WPT-OFDM) system has been proposed and compared
with its conventional counterparta Fourier transform based
optical OFDM (FT-OFDM) system. The potential deficiency
due to its double-sideband characteristics has been discussed,
and the polarization-mode dispersion (PMD) impact on
WPT-OFDM systems has been analyzed. Simulation results
show WPT-OFDM is very sensitive to PMD. Nonlinearity
performance of WPT-OFDM systems has also been studied in
this work. The result shows that the Haar wavelet has the best
tolerance and a 0.9-dB improvement in nonlinear power limit
is gained in a 2 100 km transmission setup with fully inline
dispersion compensation.
REFERENCES
[1] W. Shieh and C. Athaudage, Coherent optical orthogonal frequency
division multiplexing, Electron. Lett., vol. 42, pp. 587588, 2006.

3528

[2] A. J. Lowery, L. Du, and J. Armstrong, Orthogonal frequency division

multiplexing for adaptive dispersion compensation in long haul WDM
systems, presented at the OFC2006, Paper PDP39.
[3] I. B. Djordjevic and B. Vasic, Orthogonal frequency division multiplexing for high-speed optical transmission, Opt. Exp., vol. 14, pp.
37673775, 2006.
[4] Y. Ma, Q. Yang, Y. Tang, S. Chen, and W. Shieh, 1-Tb/s per channel
coherent optical OFDM transmission with subwavelength bandwidth
access, in Proc. OFC, San Diego, CA, 2009, Paper PDP C1.
[5] R. Dischler and F. Buchali, Transmission of 1.2 Tb/s continuous
waveband PDM OFDM FDM signal with spectral efficiency of 3.3
bit/s/Hz over 400 km of SSMF, in Proc. OFC, San Diego, CA, 2009,
Paper PDP C2.
[6] A. Cohen and I. Daubechies, On the instability of arbitrary biorthogonal wavelet packets, SIAM J. Math. Anal., pp. 13401354, 1993.
[7] R. Coifman and Y. Meyer, Orthonormal Wave Packet Bases, Dept.
Math., Yale Univ., Technical Report, 1990.
[8] R. Coifman, Y. Meyer, S. Quake, and M. V. Wickerhauser, Signal
processing and compression with wavelet packets, NATO ASI Series
C Math. Phys. Sci., vol. 442, p. 363, 1994.
[9] A. N. Akansu and R. A. Haddad, Multiresolution Signal Decomposition: Transforms, Subbands, and Wavelets, 2nd ed. Newark, NJ: New
Jersey Institute of Technology, 2001.
[10] G. Strang and T. Nguyen, Wavelets and Filter Banks. Wellesley, MA:
Wellesley-Cambridge Press, 1996.
[11] C. S. Burrus, R. A. Gopinath, and H. Guo, Introduction to Wavelets and
Wavelet Transforms. Englewood Cliffs, NJ: Prentice Hall, 1998.
[12] Y. Meyer, Wavelets-algorithms and applications, Society for Industrial and Applied Mathematics Translation, 1993.
[13] I. Daubechies, Ten Lectures on Wavelets. Philadelphia, PA: SIAM
Publications, 1992.
[14] J. D. Johnston, A filter family designed for use in quadrature mirror
filter banks, in Proc. ICASSP80 Conf., 1980, pp. 291294.
[15] N. Erdol, F. Bao, and Z. Chen, Wavelet modulation: A prototype for
digital communication systems, in Proc. IEEE Southcon Conf., 1995,
pp. 168171.
[16] I. W. Selesnick, R. G. Baraniuk, and N. G. Kingsbury, The dual-tree
wavelet transform, IEEE Signal Process. Mag., vol. 22, pp. 123151,
2005.
[17] R. T. Ogden, Essential Wavelets for Statistical Applications and Data
Analysis. Boston, MA: Birkhauser, 1997.
[18] . Bulakci, M. Schuster, C. Bunge, B. Spinnler, and N. Hanik,
Wavelet transform based optical OFDM, presented at the OFC2009,
Paper OTUO6.
[19] Y. Tang and W. Shieh, Coherent optical OFDM transmission up to 1
Tb/s per channel, J. Lightw. Technol., vol. 27, pp. 35112517, 2009.
[20] S. L. Jansen, I. Morita, T. C. W. Schenk, and H. Tanaka, 121.9-Gb/s
PDM-OFDM transmission with 2-b/s/Hz spectral efficiency over 1000
km of SSMF, J. Lightw. Technol., vol. 27, pp. 177188, 2009.
[21] H. Sun, K. T. Wu, and K. Roberts, Real-time measurements of a 40
Gb/s coherent system, Opt. Exp., vol. 16, pp. 873879, 2008.
[22] X. Liu and F. Buchali, Intra-symbol frequency-domain averaging
based channel estimation for coherent optical OFDM, Opt. Exp., vol.
16, pp. 2194421957, 2008.

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 28, NO. 24, DECEMBER 15, 2010

[23] E. Yamada, A. Sano, and H. Masuda, 1 Tb/s (111 Gb/s/ch 10 ch)

no-guard-interval CO-OFDM transmission over 2100 km DSF, presented at the OECC2008, Paper PDP6.
[24] S. J. Savory, G. Gavioli, R. I. Killey, and P. Bayvel, Electronic compensation of chromatic dispersion using a digital coherent receiver,
Opt. Exp., vol. 15, pp. 21202126, 2007.
[25] W. Shieh, X. Yi, Y. Ma, and Y. Tang, Theoretical and experimental
study on PMD-supported transmission using polarization diversity in
coherent optical OFDM systems, Opt. Exp., vol. 15, pp. 99369947,
2007.
[26] Y. Tang, W. Shieh, X. Yi, and R. Evans, Optimal design for RF-tooptical up-converter in coherent optical OFDM system, IEEE Photon.
Technol. Lett., vol. 19, pp. 483485, 2007.

An Li was born in Wuhan, China. He received the B.E. degree in optical information science and technology in 2003, and the M.E. degree in physical electronics in 2006, both from Huazhong University of Science and Technology,
Wuhan, China. He is currently working toward the Ph.D. degree in electrical
and electronic engineering at the University of Melbourne, Melbourne, Australia.
His current research interests include signal processing, wavelet analysis and
optical OFDM system design.

William Shieh (S96M96) received the M.S. degree in electrical engineering-communications and the Ph.D. degree in physics from the University
of Southern California, Los Angles, in 1994 and 1996, respectively.
Since 2004, he has been with the Department of Electrical and Electronic Engineering, University of Melbourne, Melbourne, Australia. His current research
interests include OFDM techniques in both wireless and optical communications, coherent optical communication systems, and optical packet switching.
He has published more than 110 journal and conference papers, and submitted
14 U.S. patents (nine issued) covering areas of polarization controller, wavelength stabilization in WDM systems, and Raman amplifier-based systems and
subsystems.

Rodney S. Tucker (M76SM81F89) received the B.E. and Ph.D. degrees

from the University of Melbourne, Melbourne, Australia, in 1969 and 1975,
respectively.
He is a Laureate Professor with the University of Melbourne and Research
Director of the ARC Special Research Centre for Ultra-Broadband Information
Networks. He has held positions at the University of Queensland, the University of California, Berkeley, Cornell University, Plessey Research, AT&T Bell
Laboratories, Hewlett Packard Laboratories and Agilent Technologies.
Prof. Tucker is a Fellow of the Australian Academy of Science and a Fellow
of the Australian Academy of Technological Sciences and Engineering.