You are on page 1of 3

Direct-Detection Optical OFDM Superchannel

Wei-Ren Peng, Hidenori Takahashi, Itsuro Morita, and Hideaki Tanaka


KDDI R&D Laboratories Inc., 2-1-15 Ohara, Fujimino-shi, Saitama, 356-8502 Japan Email: pe-weiren@kddilabs.jp

Abstract: We will review the transmission and signal processing techniques (phase noise & IQ
imbalance compensation) of our previously-proposed direct-detection optical OFDM super-
channel (DDO-OFDM-S) over dispersion-unmanaged standard single mode fiber (SSMF) link.
OCIS codes: (060.2330) Fiber optics communications; (060.4080) Modulation.
1. Introduction
Extensive reports on direct-detection transmissions have continually been delivered for many years due to its use
of the direct and simple receiver which generally, unlike the coherent detection that needs a local oscillator and an
optical hybrid, requires only photodiodes for detection [1-2]. Most direct-detection researches had relied heavily on
the dispersion-managed links until recently that the electronic equalization has been introduced into optical
communications for recovering the chromatic dispersion (CD)-caused distortions [3]. The introduction of electronic
equalization has indeed launched a new era of dispersion-unmanaged links which discard the need for dispersion
map design and offer a lower-nonlinear transmission medium [4]. Facing this link evolution, the joint operation
between the direct-detection receiving and the newly-considered dispersion-unmanaged link should be an interesting
research topic in consideration of the implementation cost.
To incorporate the direct-detection scheme in a dispersion-unmanaged link, there have mainly been two
proposals: 1) single-carrier transmission with pre-equalization [3] and 2) multi-carrier transmission with post-
equalization [5]. The first pre-equalization proposal [3] could work very well against CD, while it requires the link
information (i.e. the exact information of the accumulated CD) beforehand which would exclude its application in
some dynamic routing networks. The second proposal [5], commonly referred to as direct-detection optical
orthogonal-frequency-division-multiplexing (DDO-OFDM), can automatically estimate and compensate CD at the
receiver without the exact link information at the transmitter. Therefore, DDO-OFDM could support both the point-
to-point and dynamic routing networks, and should be seriously repositioned due to its specific transparency to
different kinds of network architectures.
In spite of the above-mentioned advantages, the conventional DDO-OFDM has a limited transmission
performance which might restrict its application only to the short-reach networks [6-7]. Among the many proposed
DDO-OFDM systems, the self-coherent OFDM [8], at the cost of increasing the receivers complexity, has been
developed to extend both the capacity and reach of the conventional DDO-OFDM while it diminishes the inherent
advantages of using direct detection. Therefore, a laudable goal for DDO-OFDM system would be to improve its
transmission performance with simple solutions which could still keep the low-cost nature of direct detection.
Recently, we had proposed and demonstrated a DDO-OFDM super-channel (DDO-OFDM-S) with a very simple
optical multiband receiving (OMBR) method [9, 10]. OMBR utilizes the double pass-band filters (DPFs) to detect
one band per time for the multi-band DDO-OFDM-S, which not only improves the receiving sensitivity but also
enables higher spectral efficiency (i.e. through gap reduction). In this paper, we will review our DDO-OFDM-S
transmissions that had achieved the following records in the dispersion-unmanaged links that use direct detection: 1)
214 Gb/s over 720-km SSMF, 2) 117 Gb/s over 1200-km SSMF with only 1-dB OSNR penalty. We also propose a
slight-overhead (for maximum data throughput) IQ imbalance estimation approach and specifically introduce a zero-
overhead phase noise compensation method for our DDO-OFDM-S systems.
2. Experimental Setup
In Fig. 1 we depict the experimental setups for the DDO-OFDM-S systems. The detailed setup description can


Fig. 1 Experimental setup for DDO-OFDM-S. ECL: external cavity laser, AWG: arbitrary waveform generator, MZM: Mach-Zehnder modulator,
IL: 50:100 inter-leaver, SMF: single mode fiber, DPF: double pass-band filter, OBPF: optical band-pass filter, LPF: electrical low-pass filter.
We.8.A.4.pdf 1 7/27/2011 4:37:19 PM
ECOC Technical Digest 2011 OSA
978-1-55752-932-9/11/$26.00 Optical Society of America
be found in [9, 10] and thus is omitted here. With this setup we demonstrate two capacities of (1) 214 Gb/s (190
Gb/s without overhead) with 16-QAM format and (2) 117 Gb/s (100 Gb/s without overhead) with 4-QAM format.
Here we simply emphasize several points which are lacked in our previous reports. First, at the transmitter, the
length of the optical paths between the sideband and carrier branches should be controlled as similar as possible.
Any residual length difference might introduce some incoherency between the carrier and sideband, which will lead
to the phase noise after the photodiode. Secondly, both synthesizers used in the 9-comb generators (6.5 and 19.5
GHz) are frequency-locked in order to reduce the linear crosstalk between adjacent bands. This frequency locking is
more critical for 117-Gb/s system since the bandwidth of each OFDM band is ~6.4 GHz which is closer to the 6.5-
GHz band spacing. Thirdly, the double pass-band filter (DPF) is implemented with a programmable optical filter
(Finisar Waveshaper) which limits the minimum bandwidth of each pass-band as 10 GHz. At last, the electrical IQ
demodulator is assembled with two physical double balanced mixers (rather than an integrated IQ mixer), which
would unavoidably introduce some imbalances to the signal. For the phase noise and IQ imbalance issues, we will
provide the low-overhead solutions in the next section.
3. Signal Processing Method
To compensate for the IQ imbalance and phase noise, we propose a new training method for estimating
efficiently the IQ imbalances with slight-overhead and use the decision-directed phase noise compensator (DD-
PNC) to mitigate the common-phase-error (CPE) with zero-overhead [11]. They are briefly introduced as follows:
(a) Slight-Overhead IQ Imbalance Training Method:
For each frame the proposed method utilized only two consecutive training symbols: the first symbol is
randomly generated and the second symbol simply copies the first one except that it inverts all the data symbols on
negative subcarriers. Now we turn the descriptions into mathematics for easy understanding. We denote the first and
second training symbols as |o
k
, b
-k
] and |o
k
, -b
-k
], respectively, where o
k
and b
-k
represent the data symbols on
kth and kth subcarriers, respectively, and k is a positive integer ranging from 1 to N2 with N the used data
subcarrier number. Actually they are quite similar to those used for polarization de-multiplexing purpose [12]. At
the receiver, the received training symbols, disrupted by IQ imbalances, can be denoted as |p
k
, q
-k
] and |r
k
, l
-k
].
Obviously these parameters are mutually-coupled in a way expressed as follows:
j
p
k
q
-k
[ = _
E
11
E
12
E
21
E
22
_ j
o
k
b
-k
[ and j
r
k
l
-k
[ = _
E
11
E
12
E
21
E
22
_ j
o
k
-b
-k
[
The 4 elements in the 2x2 channel matrix H can then be easily derived as E
11
= (p
k
+ r
k
)(2o
k
) , E
12
= (p
k
-
r
k
)(2b
k
) , E
21
= (q
-k
+ l
-k
)(2o
k
), E
22
= (q
-k
-l
-k
)(2b
k
). After obtaining these elements the inverse matrix
of H can be derived and used for equalization. In our demonstration this method is combined with the intra-channel
frequency-domain average method [13].
(b) Zero-Overhead Phase Noise Compensation Method:
DD-PNC uses all tentative decisions of the processed OFDM symbol to estimate its common phase error (CPE),
which is later used for de-rotating this processed OFDM symbol [11]. If we ignore the noise and IQ imbalances, the
received data symbol on kth subcarrier can be simply written as
k
= uE
k
X
k
, where X
k
,
k
are the transmitted and
received symbols on kth subcarrier, respectively, E
k
is the channel response, and u is the CPE which is independent
of subcarrier index k. Assuming CPE is not that significant so that the tentative decisions are statistically reliable, we
can estimate u as u

= (1N)
k
(E
k
X
k

)
k
, where X
k

is the tentative decisions of the equalized symbols X


`
k
=

k
E
k
, and N is the number of data subcarrier. At last, the output for final decision will be
k
(u

E
k
), of which CPE
should have been greatly mitigated.
Therefore, the used processing in our DDO-OFDM-S receiving can be listed in sequence as follows: (1)
synchronization, (2) CP removal and FFT, (3) joint channel and IQ imbalance estimation using proposed training
symbols, (4) joint channel and IQ imbalance equalization with the inverse channel matrix, (5) CPE mitigation using
DD-PNC.
4. Results and Discussions
The super-channel bands from 1 to 4 are down-converted with carrier 1 and the bands from 5 to 9 are with
carrier 2. The band index is defined in the inset of Fig. 2(b). Throughout this paper the OSNR is measured with 1.6-
nm (200-GHz) resolution covering the whole signals bandwidth and later scaled to the presented value with 0.1-nm
noise bandwidth. Q factors are all derived from BER through the formula given in Fig. 2(b).
In Fig. 2 we firstly show the experimental results of 16-QAM, 214-Gb/s DDO-OFDM-S. In this experiment, we
use a high carrier-to-sideband-power-ratio (CSPR) of ~10 dB to keep the carrier in good quality after transmission.
Figure 2(a) depicts the OSNR tolerance in back to back with and without the optical carriers. The required OSNR
achieving bit error rate (BER) = 1e-3 is found to be ~34 dB due to the high carrier power. With the consideration of
We.8.A.4.pdf 2 7/27/2011 4:37:19 PM
ECOC Technical Digest 2011 OSA


Fig. 2 Experimental results of 16-QAM, 214-Gb/s DDO-OFDM-S for (a) OSNR (0.1 nm) tolerance in back-to-back. (b) Q vs. Launch power after
720-km SSMF transmission, and (c) Q vs. band index after 720-km SSMF transmission.

the utilized CSPR = ~10 dB, the require OSNR value for the central sideband, i.e. super-channel, could be estimated
to be ~24 dB, which is found to be similar to the value in a CO-OFDM system [14]. In Fig. 2(b) we show the
measured Q as a function of the fiber launch power after 720-km transmission. The optimum power is found to be
~8 dBm, which yields the best performance under the linear noise and nonlinear fiber distortion limitations. Figure
2(c) shows the results of measured Q versus band index. In the case of back to back, the 5th band shows the worst
performance caused by the insufficient bandwidth of the receiver's components; while the other 8 bands exhibit Q
2

factors higher than 12 dB. After 720-km transmission, the worst performance happens on the 1st band, which still
yields a Q higher than the 7% FEC threshold of 8.53 dB [15]. After transmission, the performance differences
among the signal bands are reduced because it turns to be the noise and distortion that dominant the performance
instead of the receiver's bandwidth.
Next we present the experimental results of 4-QAM, 117-Gb/s DDO-OFDM-S in Fig 3. The utilized CSPR is
switched to ~3 dB to minimize the required OSNR. In Fig. 3(a) we show the Q
2
factor as a function of the launch
power after 1200-km SSMF transmission. The optimum power is found to be ~2 dBm, which is smaller than the
214-Gb/s system due to the lower CSPR here. BER versus OSNR in back-to-back and after 1200-km transmission
are depicted in Fig. 3(b). We found that in back-to-back the required OSNR at BER = 1e-3 is ~21.1 dB, which
achieves a new record in the 100-Gb/s-class DDO-OFDM systems, and that after transmission the OSNR penalty is
only ~1 dB, which exhibits a similar performance when compared with a coherent system [16]. We also discuss its
PMD tolerance in Fig. 3(c). A 3-dB Q penalty is found at an instantaneous DGD value of ~10 ps, revealing that the
demonstrated system could support a PMD outage probability of lower than 7.4e-9 [10].

Fig. 3 Experimental results of 4-QAM, 117-Gb/s DDO-OFDM-S for (a) Q vs. launch power after 1200-km transmission, (b) OSNR tolerance in
back-to-back and after 1200-km SSMF transmission, (c) Q vs. 1
st
order PMD in terms of instantaneous differential group delay (DGD).
5. Conclusions
We had reviewed our recent works on DDO-OFDM-S transmission and detailed the applied signal processing
methods in the receiver. The results reveal that the DDO-OFDM-S with OMBR can exhibit a similar performance to
CO-OFDM as long as PMD has a negligible impact.

6. References
[1] P. J. Winzer et al., JLT 26, pp.3388, 2008.
[2] M. Daikoku et al., JLT 25, pp.139, 2007.
[3] R. I. Killey et al., OFC'06, Paper OWB3, 2006.
[4] S. L. Jansen et al., ECOC08, Paper Mo3E3, 2008.
[5] A. J. Lowery et al., OFC06, Paper PDP39, 2006.
[6] A. Amin et al., PTL, 22, pp.468, 2010.
[7] D. Qian et al., JLT, 28, pp.484, 2010.
[8] B. J. C. Schmidt et al., JLT, 28, pp. 328, 2010.

[9] W.-R. Peng et al., ECOC10, Paper PDP 2.5, 2010.
[10] W.-R. Peng et al., OFC11, Paper OThX1, 2011.
[11] K. Nikitopoulos et al., IEEE Tran. Comm. 53, pp. 698, 2005.
[12] X. Liu et al., JLT, 27, pp. 3632, 2009.
[13] X. Liu et al., OpEx, 16, pp. 21944, 2008.
[14] X. Liu et al., OFC08, Paper PDPC2, 2010.
[15] ITU-T Recommendation G.975.1, Appendix I.9, 2004.
[16] Q. Yang et al., JLT, 27, pp. 168, 2009.
We.8.A.4.pdf 3 7/27/2011 4:37:19 PM
ECOC Technical Digest 2011 OSA

You might also like