Feasibility Study of SOA-Based Noise Suppression for Spectral Amplitude Coded OCDMA

Abstract

We investigate the benefits of employing a saturated semiconductor optical amplifier (SOA) to reduce the optical noise in an incoherent light optical code division multiple access (OCDMA) system. In the context of spectrum slicing, SOA-based noise suppression has shown significant potential in enhancing the signal quality of noisy light. In this paper, we evaluate the viability of the technique for spectral amplitude coded OCDMA and show that the benefits of SOA-based noise suppression do not extend readily to this application due to post-SOA optical-filtering effects at the receiver. However, appreciable performance improvements can in principle be realized through optimized system and decoder design

Full text

Page 1

394JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 25, NO. 1, JANUARY 2007
Feasibility Study of SOA-Based Noise Suppression
for Spectral Amplitude Coded OCDMA
Anoma D. McCoy, Member, IEEE, Morten Ibsen, Peter Horak, Benn C. Thomsen, and David J. Richardson
Abstract—We investigate the benefits of employing a satu-
rated semiconductor optical amplifier (SOA) to reduce the optical
noise in an incoherent light optical code division multiple access
(OCDMA) system. In the context of spectrum slicing, SOA-based
noise suppression has shown significant potential in enhancing
the signal quality of noisy light. In this paper, we evaluate the
viability of the technique for spectral amplitude coded OCDMA
and show that the benefits of SOA-based noise suppression do
not extend readily to this application due to post-SOA optical-
filtering effects at the receiver. However, appreciable performance
improvements can in principle be realized through optimized
system and decoder design.
Index Terms—Optical code division multiple access (OCDMA),
optical noise, semiconductor optical amplifier (SOA).
I. INTRODUCTION
O
bandwidth and flexibility demands of local area and access
networks [1]. OCDMA allows statistical multiplexing [2]
and asynchronous operation, while giving service providers
the ability to offer service differentiation [3], highly desir-
able for last-mile end-user markets. While numerous im-
plementations of OCDMA have been investigated [1]–[6],
spectral-amplitude-coded (SAC) techniques using inexpensive
broadband sources are well suited for the cost-sensitive access
environment [6].
In CDMA, each user channel is assigned a unique code
signature, which is imprinted onto the data signal prior to trans-
mission. SAC OCDMA systems employ passive optical filters
to encode this channel signature as a sequence of spectral slices
or “chips.” At the receiver, an individual channel is selected by
performing a correlation between the incoming encoded signal
and the code sequence, such that a peak in the correlation sig-
nifies the desired channel. For accurate detection, the channel
code sequence should have a high autocorrelation while retain-
ing low cross correlation with the other interfering channels.
When used with suitable codes, balanced differential detection
in a SAC OCDMA receiver can effectively suppress multiple-
access interference (MAI) [6]. However, systems employing
PTICAL code division multiple access (OCDMA) has
been proposed as a potential solution to address the
Manuscript received July 14, 2006; revised September 12, 2006.
A. D. McCoy, M. Ibsen, P. Horak, and D. J. Richardson are with the
Optoelectronics Research Centre, University of Southampton, SO17 1BJ
Southampton, U.K.
B. C. Thomsen was with the Optoelectronics Research Centre, University of
Southampton, SO17 1BJ Southampton, U.K. He is now with the Department of
Electronic and Electrical Engineering, University College London, WC1E 7JE
London, U.K.
Digital Object Identifier 10.1109/JLT.2006.886678
broadband sources such as light-emitting diodes or amplified
spontaneousemission(ASE)sourcessufferfromintensitynoise
due to the “thermal-like” nature of incoherent light [7]. Beating
between these multiple incoherent light channels at the detector
createsadditionalnoise,whichsetstheupperlimittoachievable
system capacity [8].
Beat-noise reduction in SAC OCDMA systems has been
investigated by a number of groups [9]–[11] who have focused
on developing code families with minimal in-phase cross cor-
relation between channels in order to reduce the total amount
of interchannel beating at each detector. This is achieved by
generating codes with reduced weight (the number of “ones”
in each code sequence) relative to code length. However, at
the same received power level, lower weight spectral codes
give rise to reduced signal quality, since the intensity noise
increases inversely with channel bandwidth [7]. As such, it is
of interest to reduce this intensity noise in order to improve
the signal quality of each channel prior to transmission. The
system impact caused by the interchannel beat noise would also
be reduced in this case, and when used in conjunction with the
appropriate family of low cross-correlation codes, the system
performance could be further enhanced.
A semiconductor optical amplifier (SOA) operating in gain
saturation can be used to suppress optical intensity noise and
has been proposed for use in spectrum slicing, providing signif-
icant “intensity smoothing” of noisy light [12]. The technique
suppresses noise fluctuations over several gigahertz, while the
wide gain bandwidth of the amplifier can accommodate in-
tensity smoothing over a large optical frequency span. Addi-
tionally, the SOA has potential to be used for simultaneous
noise suppression and modulation and can be monolithically
integrated alongside other devices, allowing a compact solution
foraccessmarkets.Wethereforeinvestigatedthistechniqueasa
meanstosuppressper-channelnoiseinaSACOCDMAsystem.
Despite the benefits listed above, in the course of our re-
search, we found that the superior signal quality offered by the
saturated SOA was noticeably degraded by subsequent optical
filtering, leading to a tangential investigation into these effects
[13]. Spectral filtering at the receiver is an unavoidable part of
SAC and thus places severe limits on the partnership of the two
technologies. In this paper, we summarize our studies in this
area and discuss the feasibility of SOA-based noise suppression
for SAC OCDMA applications in light of the implications of
post-SOA filtering. Given the ramifications of this study, we
expect that a summary of our investigation would benefit other
researchers, who may also seek to extend the advantages of
SOA-based intensity noise suppression from spectrum slicing
to SAC OCDMA.
0733-8724/$25.00 © 2007 IEEE
Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on July 1, 2009 at 09:26 from IEEE Xplore. Restrictions apply.

Page 2

McCOY et al.: FEASIBILITY STUDY OF SOA-BASED NOISE SUPPRESSION FOR SAC OCDMA395
Fig. 1.
modulator. BPF: Bandpass filter. The spectral response of the code filters is given in Fig. 3.
Two-channel block diagram for SOA per-code experiments. Noise-suppression block shown in Fig. 2. PC: Polarization controller. EOM: Electrooptic
This paper is organized as follows: In Section II, we out-
line our proof-of-concept SAC OCDMA experiment where we
employ an SOA per code/channel. We compare the system
performance with and without noise suppression and discuss
the observed effects. In Section III, possible optimizations are
investigated in an attempt to reduce the filtering effects. The
conclusions of the study are given in Section IV.
II. SOA PER CODE
A. Experiment
When a saturated SOA is used to amplify a modulated
signal, the transient response of the SOA gain can produce
patterning effects that give rise to significant eye distortion [14].
Consequently, our strategy was to use the SOA to suppress the
intensity noise of an individual channel prior to modulation.
A two-channel system was implemented, largely following
that outlined in [6], but with a saturated SOA added to each
transmitter for noise suppression.
The experimental setup is shown in Fig. 1. The ASE from an
erbium-doped fiber amplifier (EDFA) is split to form the source
spectra for the two channels, which are then shaped by fiber
Bragg grating (FBG) filters Code Tx1 and Tx2 which imprint
thechannel signatureontotheopticalspectrum.TheFBGfilters
used to perform the spectral encoding and decoding operations
were fabricated in-house [15]. Each code filter consisted of
two apodized uniform gratings cascaded in series with a sep-
aration of approximately 1 mm. Encoding the signal prior to
modulation avoids pulse broadening that arises from the spatial
separation of the two grating segments in the cascaded array
[10]. As shown in Fig. 1, the encoder filters were followed by
a noise-suppression block (Fig. 2) which consisted of a booster
EDFA, a variable attenuator, a polarization controller, and an
SOA. After the SOA, each channel was modulated at 622 Mb/s
(pattern length 27− 1), using a LiNbO3electrooptic modulator.
The channel powers were balanced at the output of the second
3-dB coupler, while the bandpass filter was used to remove the
out-of-band ASE introduced by the SOA.
At the decoder, another grating filter (Code Rx1a) performs
a correlation function between the incoming signal and the
Fig. 2.Noise-suppression block.
desired user code. A second decoder filter (Code Rx1b), which
is a spatially reversed copy of Code Rx1a, is used to remove
the pulse broadening introduced by the grating cascade. The
frequency response of the reflection path matches the code
signature of the desired channel (i.e., channel 1 in this case),
while the transmission-path response is the complement of the
reflection spectrum. Thus, the output from the transmission
path represents the correlation between the received signal and
the complement of the channel code sequence. The variable
attenuator was used to balance the unmatched channel power in
both arms of the decoder, while a variable delay line was used
to equalize the path lengths and ensure simultaneous arrival
of data bits at the inputs of the 800-MHz balanced differential
receiver. The two-channel system performance (i.e., Q-factor)
was assessed using a sampling oscilloscope with a 622-Mb/s
receiver module. We also characterized the relative intensity
noise (RIN) of the continuous-wave (CW) light (i.e., bypassing
the modulators), since, unlike Q measurements, this provides
a direct assessment of the intensity noise in the optical signal
without the need to consider modulation effects. All RIN
measurements were performed at 100 MHz using a low noise
125-MHz photoreceiver and electrical spectrum analyzer at an
optical power level of −15 dBm.
The SOAs used in the noise-suppression blocks were
bulk devices from Alcatel (channel 1) and JDS Uniphase
(channel 2). The amplifiers were characterized in terms of
gain saturation in order to select a suitable operating point
for best output signal quality. An EDFA was used in each
noise-suppression block in order to provide high-enough input
power levels (∼5 dBm) to saturate the respective SOAs, and
a polarization controller was used to align the signal to the
polarization axis of the SOA for maximum noise reduction.
Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on July 1, 2009 at 09:26 from IEEE Xplore. Restrictions apply.

Page 3

396JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 25, NO. 1, JANUARY 2007
Fig. 3.
channel {0,1,0,0,1,1,1}: (a) Code Tx1 in reflection, (b) Code Rx1a in
reflection, (c) Code Rx1b in reflection, and (d) Code Rx1a in transmission.
Unmatched channel {1,1,1,0,1,0,0}: (e) Code Tx2 in reflection. The place-
ment of these gratings within the system is shown in Fig. 1.
Grating spectra for matched and unmatched code gratings. Matched
The signature codes used in this paper are based on cyclic
shifts of a single seven-chip m-sequence, {0,1,0,0,1,1,1},
which has the requisite correlation properties for perfect
channel orthogonality when used in conjunction with bal-
anced detection [6]. Each code has a weight of 4, with
two chips overlapping between each pair of users. Code 1
{0,1,0,0,1,1,1} was used for the subject channel while
Code 2 {1,1,1,0,1,0,0} formed the interfering channel,
which are also referred to as the matched and unmatched
channels, respectively.
The FBG code filters were specified with a chip spacing
of 0.25 nm and a 3-dB bandwidth of 0.23 nm, giving two
passbands of approximately 0.23 and 0.7 nm with identical
reflectivities of ∼99.5% for the two cascaded gratings. The
measured spectral response of the code filters is given in
Fig. 3. Each grating array was placed in a tunable mount which
enabled strain tuning of the code filters, facilitating wavelength
alignment between the encoder and decoder gratings. The tun-
able mount did not, however, permit tuning of individual chips
in the code.
B. Results and Discussion
Our goal in this experiment was to identify the largest
performance improvement that could be achieved using SOA-
based noise suppression. As mentioned above, this meant high
power and careful polarization control of the SOA-input light.
Measured eye diagrams at the output of the differential receiver
at a bit rate of 622 Mb/s are given in Fig. 4. When transmit-
ting the matched channel only, a peak Q of 7.2 and 5.5 was
achieved with and without noise suppression, respectively. For
the matched code, maximum power transfer occurs through
the reflection path of the decoder, while the transmission path
blocks most of the light. The power at the transmission output
was 8 dB less than the reflection path, allowing for a clear open
eye at the receiver.
For the unmatched channel, the reflection and transmission
paths each block two chips from the code, giving approximately
equal power at both inputs to the balanced detector. Fig. 4(c)
and (d) show complete eye closure when only the unmatched
channel is transmitted, indicating that cross-talk power from
channel 2 is well suppressed. Note, however, that the intensity
noise on both of the “closed” eye diagrams shows no apprecia-
ble difference, indicating that intensity noise was not effectively
suppressed in this case. We also see that when both channels are
transmitted simultaneously, the noise suppression does not pro-
duce any noticeable advantage; the eye quality degrades notice-
ably over the single matched channel, and Q approaches that of
the two-channel system with no noise suppression (Q = 3.7).
We also observed that the saturated SOA introduced substan-
tial distortion to the encoded spectrum, as shown in Figs. 5
and 6. The signal spectrum is broadened, and the peak spectral
density is shifted toward longer wavelengths (a red shift).
Subsequently, this distorted spectrum is filtered by the decoder
gratings at the receiver. Note that the unmatched channel
experiences much greater spectral alteration than the matched
channel, as in this case, half of the encoded spectrum is filtered
out in each arm of the differential receiver [Fig. 6(b)].
As the main difference between the two received channels
is the extent of decoder filtering, the eye diagrams above sug-
gested that the signal quality of the “intensity-smoothed” light
was degraded by optical filtering. Despite the effective mitiga-
tion of MAI through balanced detection, the spectral filtering
at the decoder increases the noise in the unmatched channel,
rendering it comparable in quality to the original signal without
noise suppression. As in the pure thermal-light scenario, system
performance is limited by beat noise at the detector.
These filtering effects are more clearly illustrated using
a simple characterization experiment [see Fig. 7(a)]. In this
case, the “encoding” filter is a simple spectral slice (0.24-nm
3-dB bandwidth), as given in Fig. 7(b). After noise suppression
by the saturated SOA, the broadened slice is optically filtered
at the receiver with “decoder” filters of varying bandwidths.
Q measurements at 622 Mb/s are given in Fig. 7(c) as a function
of the post-SOA decoding filter bandwidth. All measurements
were taken at the same power levels into the detector. These
measurements clearly show that the system performance is
a strong function of the filter bandwidth and that the signal
quality declines sharply as the SOA-output spectrum is altered
by filtering. RIN measurements on the CW signal also show
Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on July 1, 2009 at 09:26 from IEEE Xplore. Restrictions apply.

Page 4

McCOY et al.: FEASIBILITY STUDY OF SOA-BASED NOISE SUPPRESSION FOR SAC OCDMA397
Fig. 4.
noise suppression for a single unmatched channel, and (e) without and (f) with noise suppression for the 622-Mb/s two-channel system. For the matched channel,
the SOA improves the Q from 5.5 to 7.2, while no change is observed for the two-channel system (Q = 3.7).
SOA per-code experiments: Eye diagrams are shown for (a) without and (b) with noise suppression for a single matched channel, (c) without and (d) with
Fig. 5.
solid), and transmission path (gray dashes) for (a) matched channel only
without SOA and (b) matched channel only with SOA.
Spectra at the decoder input (dots), decoder reflection path (black
Fig. 6.
solid), and transmission path (gray dashes) for (a) unmatched channel only
without SOA and (b) unmatched channel only with SOA.
Spectra at the decoder input (dots), decoder reflection path (black
the same trend. A comprehensive study into post-SOA spectral
filtering effects is given in [13].
Returning to the SAC OCDMA experiment of Fig. 1, the
measured RIN values at the SOA input, output, and decoder
path outputs are shown in Table I. Simulation results for
channel 1 are also given and were obtained using a traveling-
wave SOA model based on the amplifier carrier density and
field equations as described in [16]. As expected, simulation
values show good agreement with the experiment. It is clear
that the saturated SOA provides a substantial reduction in the
intensity noise of the thermal-light input, improving the signal
quality by ∼20 dB. The intensity noise level noticeably in-
creases after the decoder due to the post-SOA filtering effects
discussed above. The matched-channel noise contribution from
the transmission path is negligible due to the very low power
level. However, the noise on the unmatched channel is the
cumulative noise from both detectors of the differential re-
ceiver, and thus, the saturated SOA provides little benefit in
this case, as is apparent from the measured RIN. Also, note
that although the unmatched channel power is evenly balanced
between the two inputs of the receiver, the degree of filtering is
somewhat unequal. The imbalance arises from the asymmetric
red shift introduced by the saturated SOA, as well as the slight
mismatch in filter response between the encoding and decoding
gratings; this explains the observed difference in RIN between
the decoder reflection and transmission paths for the unmatched
channel. The observed spectral distortion of the transmitted
signal could in principal be avoided by moving the noise-
suppression block ahead of the encoder. However, any noise
suppression provided by the SOA in this configuration would
be diminished by the encoder-filtering process, providing neg-
ligible overall benefit.
Qualitatively, the signal degradation caused by post-SOA
optical filtering can be explained by considering the nonlinear
interactions that take place within the saturated amplifier. The
gain compression that occurs in saturation acts across the entire
gainbandwidthoftheSOAandintroducescorrelationsbetween
the intensity noise of the various spectral components propagat-
ing through the amplifier. We have shown both experimentally
and via simulations [13] that in distinct contrast to thermal
light (e.g., the SOA-input light), anticorrelations exist between
the intensity noise of different spectral components, which in
turn yield reduced fluctuations in the total output intensity
(i.e., noise suppression). Subsequent optical filtering removes
frequency components that contribute to the intensity smooth-
ing, thereby counteracting the noise suppression introduced by
the saturated SOA.
Therefore,theachievablenoisesuppressionintheunmatched
channel is severely limited by the optical filtering inherent
Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on July 1, 2009 at 09:26 from IEEE Xplore. Restrictions apply.

Page 5

398JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 25, NO. 1, JANUARY 2007
Fig. 7.
SOA output (dashed). (c) Q at 622 Mb/s as a function of post-SOA filter bandwidth. Decoder filters are of the same spectral shape as the encoding filter. The input
power to the detector was held constant for all filters.
(a) Single channel performance characterization with simple encoding and decoding filters. (b) Encoder-filter spectral response (solid) and broadened
TABLE I
SINGLE-CHANNEL RIN FOR SOA PER-CODE EXPERIMENTS
in the decoding process. In a practical multichannel SAC
OCDMA system, the total noise contribution from the many
interfering channels would outweigh the improvement to the
matched channel, minimizing the advantage of this approach.
Furthermore, any phenomenon that alters the correlations
between the spectral components of the intensity-smoothed
light (e.g., dispersion) can also lead to additional signal
degradation [13], [17].
III. MINIMIZING SPECTRAL FILTERING
It is clear that if SOA-based noise suppression is to provide
a significant performance enhancement in SAC OCDMA ap-
plications, post-SOA spectral filtering effects must be greatly
reduced. This can be accomplished at the system level by
restructuring the encoder to employ an individual SOA for
each spectral chip and then composing the code sequences by
coupling together the intensity-smoothed chips. In this con-
figuration, the decoding process does not drastically alter the
distinct output spectrum of any individual SOA. Clearly, as
an engineering solution, this is impractical for any realistic
code length and channel count. However, our goal was to
assess the upper limits on the performance of SOA-based noise
suppression in SAC OCDMA systems. In keeping with this, a
brief study was undertaken as described below.
For this investigation, the code sequence and decoder config-
urationwereretainedfromthepreviousexperiment(SectionII),
while new encoding filters were required at the transmitter. The
encoder gratings were fabricated as “single-chip” grating struc-
tures, with a 3-dB bandwidth of 0.24 nm and spectral response
as given in Fig. 7(b) (a representative SOA-output spectrum is
also shown). After intensity smoothing by individual SOAs, the
encoder chips at the different wavelengths were combined to
form the two separate codes; the chip spacing was 0.25 nm
as before. The channels were then individually modulated at
622 Mb/s, after which they were multiplexed together using
a 3-dB coupler prior to transmission. The decoded signal was
assessed using a sampling oscilloscope and a 622-Mb/s receiver
module. As before, the operating point of each SOA was
optimized for best noise suppression. In this system configu-
ration, the input polarization of each SOA had to be optimized
concurrently with balancing the power and modulation depth
across all spectral chips to obtain an accurate evaluation of the
two-channel system performance.
We found that the intensity fluctuations on both the ones
and zeros were observably lower when incorporating the SOAs,
and Q increased from 3.7 to ∼5 relative to the SOA per code
system. This improvement is attributed to the reduced post-
SOA spectral filtering achieved by using an individual SOA for
each chip.
However, to obtain the maximum benefit from the noise
suppression, the frequency response of the decoder must ac-
commodate the spectral broadening and distortion introduced
by the SOA. Increasing the passband of the decoder and
aligning it with the red-shifted SOA-output spectrum, together
with increasing the interchip spacing of the transmitted code,
will minimize the post-SOA filtering at the receiver. The ideal
decoder grating must also maintain high reflectivity in order
to provide adequate extinction of the undesired chips in trans-
mission and ensure the effectiveness of balanced detection.
Authorized licensed use limited to: UNIVERSITY OF SOUTHAMPTON. Downloaded on July 1, 2009 at 09:26 from IEEE Xplore. Restrictions apply.