## Abstract

This paper proposes a novel bandwidth-elastic and energy-efficient passive optical network (PON) based on the coherent interleaved frequency division multiple access (IFDMA) scheme. We experimentally demonstrate the coherent IFDMA-PON uplink transmission up-to 30 Gbps over a 30 km standard single-mode fiber with 2 × optical network units (ONUs). A low-complexity digital carrier synchronization technique enables multiple access of the ONUs on the basis of 78.1 MHz narrow band orthogonal subcarriers without any guard-bands.

© 2013 OSA

## 1. Introduction

Flexibility in the bandwidth resource usage is a vital concern for the development of next-generation optical access networks. Next-generation passive optical networks (PONs) will consist of several (e.g., 128–1024 [1]) heterogeneous optical network units (ONUs) within a coverage radius of more than 60km. Each ONU supports a different service such as legacy T1/ E1, Ethernet, RF mobile back-haul, IPTV, VPN, and upcoming machine-to-machine communications. These services have different bandwidth and latency requirements. Moreover, the requirements may dynamically change with time. For instance, hundreds of subscribers in a residential area will require a relatively small bandwidth of around 1Gbps or even smaller per subscriber and more bandwidth will be used at nights, while a business sector may require large bandwidths as high as 100 Gbps, but only in the daytime. Meanwhile, the 4G mobile back-haul may simultaneously require around 100 Mbps links for each of the more than 200 cell sites simultaneously within the PON coverage area and with a the latency of less than 1 ms [2]. To meet the wide variety of requirements in a cost- and energy-efficient manner, elastic and real-time bandwidth assignments could be an essential functionality in future PONs. In addition, the improvement of the spectrum efficiency in multiple access is required, rather than that in each user’s band, because the overhead of the guard-band and/or guard interval becomes dominant when hundreds of users coexist. Current PONs based on a time division multiple access (TDMA) need complex scheduling to support such a variety of services, and their performance is highly sensitive to other data traffic on the same link. On the contrary, a wavelength division multiple access PON (WDM-PON) has the capability to accommodate a large volume of traffic and to support multiple services due to its point-to-point topology. However, a high-spectral efficiency is difficult to realize because the data granularity is a unit of wavelength, and it requires guard-bands.

Recently, orthogonal frequency division multiple access (OFDMA), which is widely employed for physical layer multiple access schemes in mobile uplinks such as 3GPP-LTE [3] and mobile WiMAX [4], has emerged as an attractive candidate for the multiple access scheme in the next-generation PONs [5–7]. In OFDMA-based PONs (Fig. 1), the available bandwidth on a single wavelength is divided into hundreds or thousands of orthogonal subcarriers, and a different set of subcarriers are dynamically assigned to each user in each time-slot depending on the bandwidth requirement of each ONU. The adaptive assignment of numerous digitally controlled subcarriers guarantees the fine granularity and extreme flexibility in the bandwidth usage. Moreover, by using the orthogonal subcarrier as a basis of orthogonality between users, OFDMA essentially enables frequency domain multiple access without any guard-band. So far, several OFDMA-PONs have been proposed pursuing the bandwidth-elastic PON [8–10] (see also [6]); however, few of them can fully exploit the flexibility of the OFDMA concept in their uplink transmission. In the conventional OFDMA-PONs, the cost-effective intensity modulation and direct detection (IM/DD) scenario is mainly employed for their upstream traffic, and each ONU would still require a different wavelength to avoid broadband-beating noise at the optical line terminal (OLT) receiver [10]. Consequently, the bandwidth granularity of the IM/DD-OFDMA-PONs retains the same degree of WDM-PONs. A fully bandwidth-elastic PON based on the OFDMA concept can be realized by the so-called *coherent* OFDMA-PON architecture, where multiple optical in- and quadrature- phase (IQ) modulators at ONUs and a coherent receiver at the OLT enable the optical baseband OFDMA on a single wavelength. Until now, due to the costly state-of-the-art coherent transceivers and the difficulty encountered in synchronizing multiple local oscillators (LOs) at ONUs, few studies have been done on the coherent scenario, although it has the potential to achieve two or more times the spectrum efficiency than any IM/DDs [11]. However, because of the recent development of the optical coherent communication techniques, the feasibility of the coherent-PON has been demonstrated [12, 13]. In addition, some advanced OFDMA-PONs have employed coherent OLTs to improve their capacity or reach [14, 15]. This strongly motivates us to tackle the realization of the true OFDMA concept based upon a digital coherent technology in PONs.

In this paper, as one of the coherent OFDMA-PONs, we propose a digital coherent interleaved FDMA-PON (IFDMA-PON) for future elastic and power-efficient PONs. First, we introduce a novel digitally supported PON architecture for the coherent OFDMA-PON. One major challenge in coherent OFDMA-PONs is the carrier synchronization of free-running lasers at the ONUs. In optical coherent multipoint-to-point communications, the carrier frequency offset (CFO) between multiple lasers at ONUs will introduce a collision between users’ subcarriers, which will be observed as a multi-user interference (MUI) at the OLT. To overcome the MUI problem, we employ cooperative DSPs for the CFO precompensation at each ONU and the MUI cancellation at the OLT. The advanced DSPs enable coherent multiple access based on narrow band orthogonal subcarriers, whose spectrum spacing is typically less than 100MHz. Then, we discuss the advantage of the IFDMA scheme in the coherent OFDMA-based PONs. Among the OFDMA schemes, IFDMA is known for having the lowest peak-to-average power ratio (PAPR) of the modulated signal and for its simplicity in the subcarrier generation process without using the discrete Fourier transform (DFT) [16, 17]. In the proposed architecture, these characteristics are fully exploited to improve both optical and electrical power efficiencies at mass ONUs. The power consumption of DSP circuits is one concern in DSP-enabled PONs [18, 19], and the digital subcarrier generation realized using DFT, which is commonly implemented by a fast Fourier transform (FFT) circuit with multi-gigabit per second throughput, is a notable example of such a power hungry DSP in OFDMA-PONs. By removing FFT circuits from mass ONUs, the IFDMA-PON can improve the electrical power efficiency of OFDMA based PONs. In fact, the subcarrier generation without DFT in IFDMA achieved a 93% reduction of the power consumption compared with that in raw OFDMA [20]. In addition, the lower PAPR of the IFDMA signal enables a 6 dB higher optical power efficiency than other OFDMA signals in the electro-optic (E/O) conversion using Mach–Zehnder IQ modulators.

To investigate the feasibility, the IFDMA-PON uplink transmission with 2 × ONUs is experimentally demonstrated. First, we evaluate the statistical performance of the proposed CFO precompensation and show that it is possible to suppress the CFO between the two free-running external cavity lasers (ECLs), which have an initial CFO of around ±200 MHz, and then goes down to ±5MHz. By jointly using the ONU side precompensation and the OLT side interference cancellation techniques, the proposed digital coherent PON architecture realizes the coherent multiple access by using 128 subcarriers with a subcarrier spacing of 78.1 MHz without any guard-band between them. We evaluate the bit error rate (BER) performances of a 30 Gbps IFDMA-PON in the current 10G-PON scenario, i.e., a 20km standard single-mode fiber (SSFM) linked with a 1:16 power splitter, and confirm the error-free operation with a 7% forward error correction (FEC). Moreover, the elastic bandwidth assignment for the two ONUs with 10 km different link distances, i.e., 20 km for a 7.5 Gbps ONU and 30 km for a 15Gbps ONU, is also demonstrated. To the best of the our knowledge, this is the first demonstration of no guard-band and bandwidth-elastic optical coherent multiple access based on OFDMA.

The remaining of this paper is organized as follows; Section 2 presents the CFO- induced MUI problem in the coherent OFDMA-PON and proposes the digital coherent PON architecture. In Section 3, we discuss the advantage of IFDMA among the OFDMA-based PONs and show how it can be a green solution. The experimental results are shown in Section 4. Section 5 summarizes and concludes this paper.

## 2. Digital coherent OFDMA-PON architecture

One major challenge encountered when realizing the coherent OFDMA-PON is the synchronization of
multiple ONUs. Due to the lack of readily available optical phase lock loop circuits, the mitigation
of the CFO between laser sources at transceivers is a primal task for optical coherent
communications. In so-called *digital* coherent architecture, a point-to-point
coherent communication is established by compensating the CFO via the receiver side DSP. In coherent
multiple access systems, there may be serious problems with both the CFO between transceivers (i.e.,
an ONU and the OLT) and also the CFO between transmitters (i.e., ONUs). The Fig. 2 depicts the MUI problem due to the CFO between ONUs in the coherent
OFDMA-based PON. When the LOs at ONUs are freely running, the independently shifted users’
subcarriers may collide with each other after the power splitter, and the superposition of
users’ subcarriers will be observed as the MUI at the OLT. Clearly, the CFO compensation at
the ONU cannot resolve the MUI. It should be mentioned that the MUI is detrimental in the
IM/DD-OFDMA PONs. The multiple randomly shifted carriers and the users’ subcarriers,
themselves already interfered with each other, will be intermixed by the square law detection at the
OLT. To avoid the serious MUI, the conventional OFDM(A)-PONs based on the IM/DD are required to
introduce the guard-bands between user’s subcarriers, or to assign a different wavelength
for each ONU. As a result, the spectrum efficiency and the flexibility of the bandwidth usage in the
conventional IM/DD OFDM(A)-PON uplinks retain the same degree as that of WDM-PONs or FDMA-PONs.

To overcome the MUI problem, we propose a simple digitally enabled coherent OFDMA-PON architecture, as in Fig. 3. The key features in the architecture are the digital CFO precompensation at each ONU based on the channel state information (CSI) feedback and the OLT side DSP for the MUI cancellation. As an a analogy to the digital CFO compensation in the *digital* coherent receiver, the CFO precompensation is performed by multiplying the complex sinusoidal in the digital domain before E/O IQ modulation. Note that the CFO precompensation does not tune the laser wavelength but the allocation of the signal spectrum. The digital spectrum shifting is somewhat simple, but reasonable, because it enables sub-megahertz order precise frequency synchronization of the ONUs, which is not easily achieved by tuning the laser wavelength. The CFO parameter for the precompensation is estimated at the OLT, and each ONU accesses the CSI through the feedback path. The CSI feedback procedure is essentially the same as the granting and reporting procedure in current TDMA-PON standards [21], and hence it might not lose the feasibility of the proposed precompensation technique significantly. For instance, in current E-PON, the MPCPDU (multipoint control protocol data unit) consists of 64 byte [21] and the periodicity of the granting and reporting is order of tens of milliseconds typically [22], while the 8 byte CFO information feedback per ONU is required once a few tens of seconds in our demonstration will be shown in Sec. 4.

One drawback of the digital CFO precompensation is its limited tuning range. The modulation of the frequency-shifted signal spectrum will require an extra electrical bandwidth for the analog front end, such as a digital-to-analog converter (DAC). In current optical OFDM systems with 10–40 Gsample/s DACs, it may be inefficient to cancel the CFO of several gigahertz in the digital domain. With respect to CFOs larger than 1 GHz, some remote optical carrier synchronization techniques may be useful [23], where we discussed the possibility of the remote synchronization of the lasers at ONU and OLT over 100 km link numerically. Another drawback of the CFO precompensation is its poor CFO-tracking performance due to the feedback delay. To overcome this, we also introduce the digital MUI cancellation technique at the OLT side, as shown in Fig. 3.

The MUI cancellation techniques have been widely studied for wireless OFDMA systems to mitigate
the carrier synchronization errors directly at the base station (See the comprehensive survey in
[24]). They may be divided into two classes;
one is iterative interference cancellation (IC) based algorithm, another bases on linear multiuser
detection (LMD) algorithm. The typical example of the IC-based algorithm is a parallel IC-based
algorithm proposed by Huang and Lateief (HL) [25]. The basic principle of the HL method is to generate the replica of the MUI
component stemming from each user’s signal iteratively and to extract it from the received
signal so as to suppress the MUI as shown in Fig. 3.
Meanwhile a notable example for the LMD is one derived by Cao-Tureli-Yao-Honan (CTYH) in
[26], which is closely related to the
multiuser detection technique in code division multiple access (CDMA) systems. The CTYH method bases
on the optimum linear (non-iterative) decorrelator, commonly represented as a matrix, which will
regenerate the orthogonality between users’ subcarriers and minimize the MUI at the output.
The computational complexity of the HL method with *N _{itr}* (typically
< 5) iterations is given by

*O*(

*M*

^{2}

*K*

^{3}·

*N*) where

_{itr}*M*subcarriers out of

*KM*subcarriers are uniformly assigned to

*K*users. That for the CTYH method is

*O*(

*K*

^{3}

*M*

^{3}), mostly from the

*KM*×

*KM*matrix inversion. Generally, the CTYH offers better performance than the HL for larger CFOs [24, 27] at the expense of the complexity. Both techniques introduced here are applicable to any sub-carrier allocation strategies, including the interleaved (distributed) and localized (block-wise) allocations. It is worth saying that, once the system designer decides the allocation strategy, further performance improvement can be achieved by employing some advanced MUI cancellation techniques tuned specifically for the allocation, e.g., the time-domain CFO compensation technique combined with the successive MUI cancellation in the frequency domain for the interleaved allocation in [27].

The MUI cancellation methods are free from the feedback delay and can significantly simplify the task for the carrier synchronization of the ONUs. However, most of the cancellation methods can be adopted for the CFO in less than half of the subcarrier spacing. Typically, the half subcarrier spacing is less than 50MHz (39 MHz in our demonstration) in current coherent OFDM systems. Meanwhile several hundred of megahertz in the initial CFO can easily be observed even with current narrow linewidth lasers. Therefore, the cooperative operation of the pre and post CFO compensation is important for realizing the coherent OFDMA-based PONs, i.e., keep the CFO between ONUs below a half of the subcarrier spacing via the CFO precompensation at each ONU and remove the MUI due to the residual CFO at the OLT via the MUI cancellation.

Note that a larger number of subcarriers within a given electrical bandwidth provide a finer granularity and less overhead due to the cyclic prefix (CP) but a tighter acceptable CFO range. Meanwhile, finer synchronization will be achieved at the expense of the system overhead due to faster CSI feedback and the increment of computational complexity of the MUI cancellation algorithm. Therefore, a deliberate choice is required for system designers for a number of subcarriers, the CSI feedback speed and periodicity, and the MUI cancellation algorithm, for a given set of lasers at ONUs.

Finally, it is worth saying that, as for the downlink, the DD-OFDM based techniques might be attractive. In the OFDM(A) downlink, since all the subcarriers are simultaneously generated at the OLT, there is no CFO issue. Hence the IM/DD-OFDM techniques are applicable without any overhead due to the guard band. In this situation, the DD-based OFDMA schemes are reasonable because they do not require the cost-ineffective coherent receivers at math ONUs. In addition, some advanced pre-distortion technique at the OLT may be interesting for further reduction of the power-consumption as well as the cost of the DSP circuit at the ONUs in such DD-OFDM based downlinks [28].

## 3. IFDMA: Energy-efficient OFDMA without DFT

The architecture given in Fig. 3 basically enables any type of OFDMA schemes. Our next interest is therefore to determine the scheme among the OFDMA family that is suitable for PON architectures. In this section, we show how the IFDMA-PON could be a green solution for the coherent OFDMA-PON.

#### 3.1. Electrical power efficiency in digital subcarrier generation

One known drawback of the raw OFDMA scheme is the high PAPR of the modulated signal. Due to
constructive and destructive interference between hundreds or thousands of subcarriers, the
time-domain waveform of the OFDMA signal becomes like a Gaussian noise, and a peak power of 10 dB
larger than the average power, is observed in some cases. Single carrier FDMA (SC-FDMA) is a
technique that is used to reduce the PAPR of the raw OFDMA and has been widely employed in current
mobile uplinks, such as in 3GPP-LTE [3]. IFDMA
is a special class of SC-FDMA where a comb-shaped subcarrier allocation is employed. The Fig. 4(a) shows a basic schematic of subcarrier generation in
SC-FDMA in the digital domain. In the raw OFDMA, orthogonal subcarriers are efficiently generated
using an inverse FFT (IFFT) circuit. In SC-FDMA, information-bearing symbols are precoded by the
smaller size FFT before mapping on the frequency bins of the IFFT. The FFT precoding is known to
reduce the PAPR if a localized or interleaved subcarrier set is employed as the basis of FDMA. A
notable feature of IFDMA is the simplified circuit for the subcarrier generation [16]. The SC-FDMA subcarrier generation process in Fig. 4(a), i.e., FFT, interleaved subcarrier mapping, and IFFT, is
mathematically equivalent to the blockwise repetition and the user dependent phase rotation
operation shown in Fig. 4(b). Thus, it is possible to
generate subcarriers without any FFT circuits, and the computational complexity is reduced to
*O*(*KM*) from *O*(*KM*
log(*KM*)) + *O*(*M* log(*M*))
when *K* ONUs are exist and *KM*-point IFFT is used to assign
*M* subcarriers per ONU uniformly. The feature is attractive in high-speed optical
access systems in terms of power consumption.

In current PONs, 65% of the overall PON power is consumed in mass ONUs [19]. Moreover, in the future digitally enhanced PON ONUs, high-throughput DSP subsystems can be the major energy consumer [18]. For instance, in the 10 Gbps ONU in the Hybrid WDM/O-OFDM-PON, whose ONU configuration is almost the same as the ONU in the coherent OFDMA-PON, nearly 70 % of the power is consumed at the DSP circuit for the OFDM modulation/demodulation. An IFFT circuit with throughput of several gigasample per second is a notable example of such a power hungry DSP subsystem at the ONU. By removing the high-speed IFFT from the mass ONUs, the IFDMA-PON can decrease the total power consumption of OFDMA based PON systems.

In [20], we have experimentally investigated the power consumption of the FPGA (field-programmable gate array) -based 10 Gbaud OFDMA transmitter circuits. The Fig. 5(a) depicts the IFDMA transmitter by employing the static random access memory (SRAM)-based FPGA circuit, and a comparison of the power consumption of the IFDMA circuit and a conventional OFDMA circuit is shown in Fig. 5(b). In the experiment, the ALTERA Stratix IV FPGA is used. The sampling rate and operation accuracy are 30 Gsample/s and 8 bit, respectively, and the toggle rate is 12.5 %. As in Fig. 5(b), the high-throughput IFFT circuit in the 10 Gbaud OFDMA nearly consumes 100W. Meanwhile, the proposed IFDMA circuit that employs the SRAM-based circuit demonstrated up to 93% reduction in power consumption.

Note that the power dissipation of the high-speed DACs and ADCs is another concern. However, owing to the recent development of 100 Gb/s long-haul fiber transmission systems, the cost as well as energy efficiency of these high-speed DAC and ADC are significantly improving [6]. For instance, the 55–65 Gsample/s DAC with 8 bit resolution developed with the 40 nm CMOS (complementary metal-oxide-semiconductor) technology only consumes 0.75 W/ch [29]. Similarly, the power consumption of the recent 8 bit 55–65 Gsample/s ADC on the 40 nm CMOS is reported as 1.5 W/ch [29]. Thus the expected power consumptions in DACs and ADCs in the IFDMA-PON uplink are only 1.5 W per ONU Tx (two DACs for the optical IQ modulation) and 6 W for the OLT Rx (four ADCs for the polarization diversity optical coherent reception), despite of > 2 times higher sampling rate. These values are not so significant compared with the IFFT circuit shown in Fig. 5(b). The ASIC implementation of the IFFT circuit will reduce their power consumption significantly. In [30], the authors investigate the optimal 65 nm ASIC implementation of the digital OFDM modulation. The estimated power consumption for 10 Gbaud OFDM modulation with only 128 subcarriers is 0.4–0.56 W excluding the some related DSPs. The amount is still comparable to the consumption at the DAC.

#### 3.2. E/O conversion efficiency

Another power efficiency of the IFDMA-PON can be found in the E/O conversion process. Null-biased
dual-parallel Mach–Zehnder modulators (DP-MZMs) are commonly used for a linear E/O IQ
conversion in optical coherent systems employing advanced modulation formats, such as QPSK, 16QAM
and OFDM(A). A single MZM is known to have a sinusoidal transfer response, and hence the linear
region of the E/O conversion is limited for an input driving voltage smaller than the 100%
modulation voltage *V _{π}* [Fig. 6(a)], even with the ideal predistortion in the electrical domain
[31]. The linear region of the E/O IQ
modulation via the DP-MZM may be projected onto the square region on the complex baseband plane, as
shown in Fig. 6(b). As long as the baseband representation of
the electrical waveform is in the region, the waveform is linearly converted to the optical domain.
The solid square in the figure represents the 100% modulation voltage of each MZM and thus
corresponds to the maximum optical output power from each MZM. Therefore, when the baseband signal
trajectory is closer to the square, we can expect a higher modulation efficiency. This implies the
poor optical power efficiency in the E/O IQ conversion of the high-PAPR signals. The Fig. 7(a) depicts the trajectories of the baseband OFDMA signal on
the complex plane, where a rectangular pulse shaping is employed, and each point results from the
4-fold oversampling of the analog waveform. To maintain a linear E/O conversion even for the peaks
in the OFDM(A) waveform, the average input voltage should be suppressed as the solid circle in Fig. 7(a). Actually, the optical power efficiency (OPE) of the E/O
IQ conversion [31] may be related to the
average PAPR of the modulated signal, i.e.,

*α*is a constant for the E/O conversion efficiency and the instantaneous PAPR is defined as $\u3008\frac{\mathit{Maximum}\hspace{0.17em}\mathit{instantaneous}\hspace{0.17em}\mathit{power}\hspace{0.17em}\mathit{in}\hspace{0.17em}\mathit{an}\hspace{0.17em}\mathit{OFDM}\hspace{0.17em}\mathit{symbol}}{\mathit{the}\hspace{0.17em}\mathit{average}\hspace{0.17em}\mathit{power}\hspace{0.17em}\mathit{of}\hspace{0.17em}\mathit{the}\hspace{0.17em}\mathit{OFDM}\hspace{0.17em}\mathit{symbol}}\u3009$. Hence, an OFDM signal with an average PAPR of 10 dB will lose at least 90% of the laser power during the modulation process. In [31], the authors investigate the OPE of the optical OFDM signal with the optimal hard clipping. Even with the optimal clipping, the OPE of the raw OFDM(A) signal is around 12% (−9 dB). Meanwhile, Fig. 7(b) represents the trajectory of the baseband IFDMA signal with the 8PSK format. As in Fig. 4(b), the IFDMA subcarrier generation process consists of the repetition and phase rotation. Therefore, there is no factor to increase the PAPR of the modulated signal from that of the input information-bearing symbols. In particular, if we employ a PSK format where the carrier signal amplitude is constant during the phase transitions, the resulting IFDMA signal amplitude remains constant, i.e., PAPR = 0 dB, as in Fig. 7(b). Note that pulse shapes other than the rectangular pulse will increase the PAPR of the IFDMA signal [17]. The impact is significant when the shaping filter with brick-wall frequency response is employed, but it is not so common in optical communications systems.

The expected OPE for IFDMA is up to 50% (−3 dB). In PONs where the system power budget is defined by the power of the laser source and the sensitivity of the receiver, the loss of the laser power during the E/O conversion directly affects the reach of the system. The 6 dB improvement of the OPE implies that the IFDMA-PON can extend the reach by 20 km (in the case where the fiber loss = 0.3 dB/km) compared with the PON that is based on the raw OFDMA.

In addition, the IFDMA scheme also has an advantage in its nonlinear tolerance over the other OFDMA-PONs. In [32], we reported the PAPR enhancement phenomenon in the optical OFDM transmission. The high-PAPR OFDM signal is known to be susceptible to the fiber nonlinearity even if the average launched power is a few. Moreover, after tens of kilo meters propagation over SSMF, the PAPR of the optical OFDM signal is suddenly increased by more than 1 dB due to the optical Kerr effect. This increases the severity of nonlinear penalty in OFDMA-PONs relative to that in the conventional PONs. The low-PAPR IFDMA scheme is also expected to improve the nonlinear tolerance of OFDMA-PONs.

## 4. Experimental results

Here, the no guard-band and bandwidth-elastic optical coherent access of 2 × ONUs based on IFDMA is experimentally demonstrated to validate the feasibility of the proposed digital coherent IFDMA-PON system. The Fig. 8 shows the experimental setup for 2 × ONU IFDMA-PON uplink experiments including the employed offline DSPs. At the ONUs, two-independent IFDMA signals with 128 subcarriers including periodic short pilots and 11% CP are generated offline in a workstation, and each of them is analog-to-digital converted by a 10 Gsample/s arbitrary waveform generator (Tektronix AWG7122C). The AWG outputs are then E/O converted by DP-MZMs and Rio PLANEX™ External Cavity Lasers (ECL1 and 2). The wavelength of ECL1 and ECL2 is 1550.15nm, and the (electrically) fitted Gaussian FWHM (full width at half maximum) linewidths are 9.4 kHz and 10.3 kHz, respectively. Note that they are freely running after the initial tuning. In addition, to focus on the CFO between the ONUs and validate the proposed digital carrier synchronization technique, the ECL1 is also used as the LO at the OLT. The generated optical IFDMA signals were then launched into SMF-28 (SSMF1 and SSFM2). The lengths of the SSFMs will be changed by 0–20km, while no optical amplifier is employed in the setup. At the OLT, the outputs of the coherent receiver are observed using real-time sampling oscilloscope (Tektronix DSA72004C) at 50 Gsample/s. The acquired waveform information is then transferred to the workstation over TCPIP, and all the DSP, including the IFDMA demodulation, are performed offline. Note that the AWGs and the DSA are connected to the same workstation, and this enables the CSI feedback for the digital CFO precompensation. As for the MUI cancellation, we employ the HL algorithm with 5 iterations. It should also be mentioned that although the high-speed sampling oscilloscope provides the 5-fold oversampling of the symbol rate, the DSP in the OLT only exploits symbol rate samples to determine the feasibility of the practical OLT with a low-cost front end.

The Fig. 9 shows the transmitted IFDMA frame designs for ONU1 and ONU2. Each frame consists of a 109 ns preamble and a 576 ns data payload. The data payload contains 40 IFDMA symbols, and the orthogonality between the users’ payloads is achieved via the OFDMA sense. Besides, in order to acquire the ONU-by-ONU CSI including the CFO parameter, the users’ preambles should also be decomposable at the OLT. Here, to ease the task for the offline CSI estimation, the orthogonality is realized by just assigning a different time slot for each preamble, namely TDMA, as shown in Fig. 9. This will lead a huge overhead when a large number of ONUs are co-exist. However, the overhead can be avoided in practical situations by employing orthogonal sequences, such as the Zadoff–Chu sequence, as preambles and/or CSI-tracking techniques with short periodic pilots.

#### 4.1. Impact of the digital CFO precompensation

First, we verify the proposed digital CFO precompensation technique. The Fig. 10 shows the complementary cumulative distribution function (CCDF) of the estimated CFO between the ONUs and examples of the received (partial) spectrum at the OLT. In Fig. 10(a), we observe the collision of users’ subcarriers due to the CFO; the comb-shaped spectrum of ONU2 (blue) almost overlaps with the spectrum of ONU1 (red) where the estimated CFO is 43 MHz, while the subcarrier spacing is 78.1 MHz. Note that the spectrums are observed one after the other for visualization here. Statistically, from the CCDF (the red dotted line), the CFO larger than 160 MHz is observed 90% of the time without the precompensation, while the half subcarrier spacing in this experiment is 39 MHz. Therefore there is a need for some CFO compensation or reduction techniques at the ONU side.

Note that the CFO is estimated based on the preamble of each transmitted data frame. Hence the CFO here is the accumulated version of the fluctuation of the laser’s wavelength during the IFDMA frame duration. Essentially, there are two requirements for the lasers used for coherent OFDMA transmission; 1) the laser are stable as they are said to be quasi-synchronous during a given transmitted frame [33], and 2) the CFOs are less than a half of the subcarrier spacing to avoid the significant performance degradation due to the collisions between user’s subcarriers. The CCDF here directly denotes the outage probability of these two requirements.

The Fig. 10(b) shows the received spectrum with the proposed digital CFO precompensation technique, where the initial CFO is the same as in Fig. 10(a). In Fig. 10(b), the two comb-shaped spectrums are properly shifted to inter-mesh, and hence, the orthogonality between the subcarriers is restored. From the CCDF (the blue line), we observe a significant reduction of the CFO via the proposed precompensation method, e.g., the probability of observing a CFO larger than ±5 MHz is around 4%, which is sufficiently small for the coherent OFDMA transmission with 78.1 MHz subcarrier spacing. One concern is the periodicity and/or the latency required for the CSI feedback for the CFO precompensation, and further investigations using real-time front ends are necessary to clarify the requirements for the CSI feedback performance. In the experiment, the AWG and the DSA are operated in store-and-forward manner. Moreover the CFO is estimated in offline manner. Therefore the latency of the CSI feedback path results in second order and much longer than the typical round trip time in PONs. Even so, the CSI feedback of once every a few tens of seconds is sufficient to maintain the carrier synchronization owing to the narrow linewidth of the ECLs.

#### 4.2. Back-to-back sensitivity

Next, we evaluate the sensitivity of the OLT in a back-to-back configuration, i.e., SSFM1
= SSFM2 = 0 km. The Fig. 11 depicts the
received optical power versus the BER performances of each ONU, where 64 subcarriers out of 128
subcarriers are equally assigned to both ONUs, and the QPSK format is employed. The resulting
transmission rate is 10 Gbps for each user and 20 Gbps in total without the FEC and CP overhead.
Both ONUs achieve error-free transmission with 7% FEC (BER = 3.8 ×
10^{−3}[34]) for a received
power larger than −35.5 dBm. Moreover, we observe almost Gaussian-like distributions in the
received constellations for both ONUs without any imaging component between them. This means that
the MUI due to the CFO between the two free-running ECLs is successfully compensated via the ONU
side precompensation and the OLT side MUI cancellation.

In Fig. 12, we increase the bit rate by employing the 8PSK format. The resulting transmission rate is 15 Gbps for each ONU and 30 Gbps in total (24.6 Gbps with CP and 7% FEC overheads). From the figure, for a received power larger than −27.5 dBm, both ONUs achieve a 7% FEC error-free operation. In comparison with the QPSK case, we observe a 7 dB penalty in the OLT sensitivity, while the theoretical penalty is around 5 dB at the BER of 1.0 × 10^{−3}[35]. A part of the additional 2 dB penalty may be from the IQ imbalance at the DP-MZMs. As seen in the constellations in Fig. 12, the received signal has some skew between the IQ components, which is mainly due to the unstable bias voltages at the DP-MZMs. This type of IQ skew is called a transmitter IQ imbalance and is known to induce unexpected MUI at the receiver side in OFDMA systems [36].

#### 4.3. Feasibility on the 10G-PON scenario

Here we test the IFDMA-PON in a current 10G-PON scenario: a 20 km SSMF link with a 1 × 16 power splitter. Again, the modulation format is 8PSK; thus, the system bit rate is 30 Gbps, excluding the overhead due to the CP, FEC and preamble. In this case, SSMF1 is set to 20 km (SSFM2 = 0 km), and the power splitter is emulated by an attenuator just before the OLT. Due to equipment limitations, the signal wave length is not the zero-dispersion wave length of SMF-28, as in current PON uplinks. The fiber dispersion is electrically compensated via the frequency domain equalizer at the OLT side. The Fig. 13 shows the BER performances of the ONUs versus the attenuator gain and examples of the received constellations. Here the DP-MZM output powers are −4.67 dBm for ONU1 and −3.47 dBm for ONU. The 7% FEC error-free transmission is achieved for an attenuation gain of less than −15 dB. Including the 3 dB loss at the coupler, the −15 dB attenuation corresponds to the 32 splits. Although the number of ONUs is limited to 2 at this time, the IFDMA-PON has the potential to fulfill the requirement for 10G-PON. Moreover, the IFDMA-PON achieves a rate three times higher than the 10G-PON with the same optical bandwidth by employing the multi-level modulation format and removing the guard-band between the ONUs based on the OFDMA scheme.

#### 4.4. Demonstration of the bandwidth-elastic multiple access

Finally, we demonstrate the elastic bandwidth allocation in the coherent IFDMA-PON. In the
demonstration, SSFM1 and SSFM2 set to 20 km and 10 km, respectively. Due to the difference in the
access spans, ONU2 suffers from a 3 dB loss in the optical signal-to-noise power ratio (OSNR) when
compared with ONU1. To mitigate the OSNR degradation without decreasing the bit rate of ONU2, we
assign 96 subcarriers out of 128 with the QPSK format for ONU2. The remaining 32 subcarriers are
allocated for ONU1 with the 8PSK format. The resulting transmission rates are 7.5 Gbps for ONU1 and
15 Gbps for ONU2, excluding the overheads. The Fig. 14
depicts the BER performances, received constellations, and subcarrier-by-subcarrier error vector
magnitude (EVM) performances, where the optical power at the OLT input is −17.24 dBm. The
BERs for ONU1 and ONU2 are 9.7 × 10^{−4} and 1.2 ×
10^{−3}, respectively, and a 7% FEC error-free operation is observed on both
ONUs. To the best of our knowledge, this is the first demonstration of no guard-band and
bandwidth-elastic optical multiple access based on the OFDMA scheme.

Meanwhile, we observe a severe distortion on the constellations in Fig. 14. In fact, the constellations are rather broadened compared with the EVM performances. This is mostly due to the imperfection in the synchronization of the sampling timing at the OLT.

In the OFDMA-based PON, the optimal sampling timing may be different for each ONU’s signal, while the OLT is required to set a fixed sampling timing for the simultaneous reception of ONUs’ signals as an OFDMA signal. One reason for the offset between the optimum sampling timings of the ONUs is the mismatch of sampling clocks in multiple high-speed DACs at the ONUs. The other source of the offset is the difference in the propagation delay. Note that OFDMA is known for its robustness to the delay between the arrival times of users’ signals. The delay can be compensated as a part of the channel response via the frequency domain equalizer (FDE), if the delay time is a multiple of the symbol duration. Meanwhile, the residual delay, which may have a duration that is a fraction of the symbol period, will induce the offset discussed here. Because the accurate synchronization of multiple and high-speed sampling clocks over a fiber link of several tens of kilometers is not simple, the OLT should choose a certain sampling timing that maximizes the average SNR of ONUs or balance the received SNRs between ONUs. In other words, the OLT is required to implement a sub optimal matched filter for the simultaneous reception of the users’ subcarriers. In the demonstration, we employ the former strategy. However, in our setup, the received optical signal is intermittently acquired and the sampling timing chosen by the timing synchronizer varied for each observation. As a result, we observed the broadening in the constellations as shown in Fig. 14.

## 5. Numerical results

Here we show some numerical results. Our aim here is to investigate the feasibility of the IFDMA-PON uplink with > 2 ONUs with > 10 kHz linewidth lasers.

First, we obtain a heuristic model for the CFO distribution of the ONU lasers after the CFO
precompensation. Based on the experimental result in Sec. 4.1, we found a Laplace distribution is
useful to represent the statistics of the CFO Δ*f*, i.e., $\text{Pr}\left(\mathrm{\Delta}f\right)=\frac{1}{2\lambda}\text{exp}\left\{-\frac{\left|\mathrm{\Delta}f\right|}{\lambda}\right\}$, where the scale parameter *λ* = 3.1
× 10^{6}. The probability distribution function (PDF) and the CCDF of
Δ*f* are depicted in Fig. 15(a) and
Fig. 15(b), respectively. From Fig. 15(b), we can see good agreement between the Laplacian model and the
statistics acquired through the experiment at least for the first trend in CCDF. Note that the
second and the third trends in the experimental CCDF are mostly due to our equipment. An analytical
model for the CFO distribution of the remotely controlled lasers is of our next interest, however
the heuristic model may be useful for this primal investigation of the MUI issue in the IFDMA-PON
uplink with > 2 ONUs.

The Fig. 16 represents the simulation model for the
IFDMA-PON uplink. Some simulation parameters are summarized in Table 1. In the model, the ONUs are assumed to distribute uniformly within a 10 km radius,
thus the link distances span 20–30 km long. Each ONU has a fixed output power of 2 mW, and
the received-power imbalance at the OLT as well as the chromatic dispersion are mitigated via the
frequency domain equalization at the OLT. The received OSNR, defined based on the OLT input power
after the multiplexing of the users’ subcarriers, is fixed to be 17 dB (23 dB in the
electrical SNR in this case) for any situations. The central frequencies of the ONU lasers are
assumed to fluctuate independently and identically on an IFDMA frame-by-frame basis. We employ the
Laplacian model as the stochastic model of the fluctuation. We test the BER performance of the
IFDMA-PON uplink with 2–64 ONUs, where the available 128 subcarriers are equally assigned to
each ONU. For instance, for the ONU count *K* = 2, interleaved 64 subcarriers
out of 128 subcarriers are equally assigned to two ONUs, and 32 subcarriers are allocated per ONU
for *K* = 4.

The averaged BER performance of the overall ONUs versus the ONU count *K* is represented in Fig. 17. The red dashed line is with the MUI cancellation based on the HL algorithm with 5 iterations, and the red solid one is without the cancellation. First, it is worth pointing that the average BER performance improves as the ONU count *K* increasing. The collision of the neighboring subcarriers generated from different lasers is a major source of the BER degradation in the OFDMA systems with large CFOs. In the IFDMA simulation with *K* = 2 (64 interleaved subcarriers per ONU), if one of the two ONUs suffers from a large CFO, all 128 subcarriers will collide each other. Meanwhile, when *K* = 4 (32 subcarriers per ONU), only 64 subcarriers will collide when a large CFO is observed at one of 4 ONUs. In other word, for *K* = 4, all 128 subcarriers can collide when at least two of the ONUs suffer from large CFOs simultaneously, and such joint probability is much less than the probability to observe a large CFO. Strictly speaking, this statistical performance improvement depends on the outage probability of Δ*f* at each ONU. However, generally speaking, the outage probability must be less than 10^{−2} when we consider the system performance around BER = 10^{−3}. The situation is the same for *K* > 4. Therefore we can conclude that the uplink experiment with 2× ONUs in Sec.4 shows the worst performance of the IFDMA-PON in sense of the average BER, and we can expect the FEC error-free operation for the uplink with > 2 ONUs with the same equipment except the overhead for the CSI estimation.

The blue dashed and solid lines in Fig. 17 represent the average BER versus *K* with and without the MUI cancellation technique, respectively. Here we employ the Laplace distribution with *λ* = 6.2 × 10^{6} for the CFO distribution. It is not easy to relate *λ* and the laser linewidth. The laser frequency stabilization technique is different for a required linewidth, and the CFO distribution after precompensation may be different as well. However, with a simple scaling, *λ* = 6.2 × 10^{6} corresponds to the linewidth of around 40kHz. In this case, with the help of the MUI cancellation at the OLT, the FEC error-free operation can achieved for every ONU count.

These numerical results are based on our experimental result with the second-order CSI feedback. Further investigation on the MUI issue and the required laser linewidth for the IFDMA-PON uplink by using the real-time setup is of our next interest. However, the numerical results show the potential of the IFDMA-PON uplink with > 2 ONUs with lower cost lasers.

## 6. Conclusion

A novel bandwidth-elastic and power-efficient coherent PON based on an IFDMA scheme has been presented. The coherent IFDMA-PON uplink transmission via 2 × ONUs with free-running sources has been experimentally demonstrated. By enabling a high-speed and multi service platform with fine-grained elastic bandwidth allocation in an energy efficient manner, the proposed IFDMA may be an attractive solution for future PON systems.

## Acknowledgment

This work has been financially supported by the R&D program, SCOPE, funded by the Ministry of Internal Affairs and Communications (MIC), Japan (FY.2011–2013).

## References and links

**1. **J. Kani, “Next-generation PONs: An operator’s view,h” in Proc. of Euro. Conf. Optical Commun., paper 5.7.4., (2009).

**2. **N. Cvijetic, A. Tanaka, Y. K. Huang, M. Cvijetic, E. Ip, Y. Shao, and Ting Wang, “4+G mobile backhaul over OFDMA/TDMA-PON to 200 cell sites per fiber with 10Gb/s upstream burst-mode operation enabling <1ms transmission latency,” in Proc. of Opt. Fiber Commun. Conf., PDP5B.7 (2012) [CrossRef] .

**3. **3GPP TS 36.300, 3rd Generation Partnership Project Technical Specification Group Radio Access Network, “Evolved universal terrestrial radio access (E-UTRA) and evolved universal terrestrial radio access network (E-UTRAN); Overall description; Stage 2 (Release 8).”

**4. **IEEE Std. 802.16e, “Air interface for fixed and mobile broadband wireless access systems amendment for physical and medium access control layers for combined fixed and mobile operation in licensed bands,” IEEE, 2006.

**5. **W. Shieh and I. Djordjevic, *OFDM for Optical Communications* (Academic Press, 2009).

**6. **N. Cvijetic, “OFDM for next-generation optical access networks,” J. Lightwave Technol. **30** (4), 384–398 (2012) [CrossRef] .

**7. **N. Cvijetic, D. Qian, and J. Hu, “100 Gb/s optical access based on orthogonal frequency division multiplexing,” IEEE Commun. Mag. **48** (7), 70–77 (2010) [CrossRef] .

**8. **N. Cvijetic, N. Prasad, D. Qian, J. Howard, and T. Wang, “Computationally- efficient DSP-based MIMO equalization for OSNR gains in 40Gb/s OFDMA-PON,” in Proc. of Opt. Fiber Commun. Conf., paper OTuK6, (2011).

**9. **B. Charbonnier, N. Brochier, and P. Chanclou, “(O)FDMA PON over a legacy 30dB ODN,” in Proc. of Opt. Fiber Commun. Conf., paper OTuK1, (2011).

**10. **D. Qian, N. Cvijetic, J. Hu, and T. Wang, “A novel OFDMA-PON architecture with source-free ONUs for next generation optical access networks,” IEEE Photon. Technol. Lett. **21** (17), 1265–1267 (2009) [CrossRef] .

**11. **D. J. F. Barros and J. M. Kahn, “Comparison of orthogonal frequency-division multiplexing and on-off keying in amplified direct-detection single-mode fiber systems,” J. Lightwave Technol. **28** (12), 1811–1820 (2010) [CrossRef] .

**12. **S. Smolorz, H. Rohde, E. Gottwald, D. W. Smith, and A. Poustie, “Demonstration of a coherent UDWDM-PON with a real-time processing,” in Proc. of Opt. Fiber Commun. Conf., PDPD4, (2011).

**13. **D. Lavery, M. Ionescu, S. Makovejs, E. Torrengo, and S. J. Savory, “A long-reach ultra-dense 10 Gbit/s WDM-PON using a digital coherent receiver,” Opt. Express **18** (25), 25855–25860 (2010) [CrossRef] [PubMed] .

**14. **N. Cvijetic, M. Huang, E. Ip, Y. Huang, D. Qian, and T. Wang, “1.2 Tb/s symmetric WDM]OFDMA]PON over 90km straight SSMF and 1:32 passive split with digitally]selective ONUs and coherent receiver OLT,” in Proc. of Opt. Fiber Commun. Conf., PDPD7, (2011).

**15. **D. Qian, S. H. Fan, N. Cvijetic, J. Hu, and T. Wang, “64/32/16QAM-OFDM using direct-detection for 40G-OFDMA-PON downstream,” in Proc. of Opt. Fiber Commun. Conf., paper OMG4, (2011).

**16. **U. Sorger, I. De Broeck, and M. Schnell, “IFDMA – A new spread spectrum multiple access scheme,” in Proc. of IEEE International Conference on Communications , 1013–1017, (1998).

**17. **H. Myung, J. Lim, and D. Goodman, “Single carrier FDMA for uplink wireless transmission,” IEEE Vehicular Tech. Mag. **1** (3), 30–38 (2006) [CrossRef] .

**18. **OASE Deliverable D4.1: “Survey of next-generation optical access system concepts”, (Oct. 2010). (available online:www.ict-oase.eupublicfilesOASE_WP4_D4_1_29th_October_2010_v1_0.pdf)

**19. **J. Baliga, R. Ayre, K. Hinton, W. V. Sorin, and R. S. Tucker, “Energy consumption in optical IP networks,” J. Lightwave Technol. **27** (13), 2391–2403 (2009) [CrossRef] .

**20. **K. Ishii, Y Akiyama, T. Yoshida, N. Suzuki, T. Ichikawa, K. Koguchi, J. Nakagaw, T. Mizuochi, Y. Yoshida, A. Maruta, and K. Kitayama, “Low-power consumption DSP circuit design for IFDMA-based PON systems,” in Proc. of Opto-Electronics and Communications Conference , 770–771, (2011).

**21. **C. Lam, *Passive Optical Networks* (Academic Press, 2007).

**22. **ITU-T Recommendation Y.1541, (2011).

**23. **Y. Akiyama, K. Ishii, T. Yoshida, N. Suzuki, T. Ichikawa, K. Koguchi, J. Nakagawa, T. Mizuochi, Y. Yoshida, A. Maruta, and K. Kitayama, “Remote inter-carrier synchronization for subcarrier multiple access PON systems,” in Proc. of Opto-Electronics and Communications Conference , 832–833, (2011).

**24. **M. Morelli, C.-C. J. Kuo, and M.-O. Pun, “Synchronization techniques for orthogonal frequency division multiple access (OFDMA): A tutorial review,” Proceedings of the IEEE **95** (7), 1394–1427 (2007) [CrossRef] .

**25. **D. Huang and K. B. Letaief, “An interference-cancellation scheme for carrier frequency offsets correction in OFDMA systems,” IEEE Trans. Commun. **53** (7), 1155–1165 (2005) [CrossRef] .

**26. **Z. Cao, U. Tureli, Y. D. Yao, and P. Honan, “Low-Complexity orthogonal spectral signal construction for generalized OFDMA uplink with frequency synchronization errors,” IEEE Trans. Veh. Technol. **56** (3), 1143–1154 (2007) [CrossRef] .

**27. **Y. Zhu and K. B. Letaief, “CFO estimation and compensation in SC-IFDMA systems,” IEEE Trans. Wireless Commun. **9** (10), 3200–3213 (2010) [CrossRef] .

**28. **X. Chen, A. Li, D. Che, Q. Hu, Y. Wang, J. He, and W. Shieh, “High-speed fading-free direct detection for double-sideband OFDM signal via block-wise phase switching,” in Proc. of Opt. Fiber Commun. Conf., PDP5B.7, (2013).

**29. **I. Dedic, “High-speed CMOS DSP and data converters,” in Proc. of Opt. Fiber Commun. Conf., paper OTuN1, (2011).

**30. **R. Bouziane, P. Milder, R. Koutsoyannis, Y. Benlachtar, J. C. Hoe, M. Puschel, M. Glick, and R. I. Killey, “Design studies for ASIC implementations of 28 GS/s optical QPSK- and 16-QAM-OFDM transceivers,” Opt. Express **19** (21), 20857–20864 (2011) [CrossRef] [PubMed] .

**31. **D. J. F. Barros and J. M. Kahn, “Optical modulator optimization for orthogonal frequency-division multiplexing,” J. Lightwave Technol. **27** (13), 2370–2378 (2009) [CrossRef] .

**32. **Y. Yoshida, A. Maruta, and K. Kitayama, “On the peak-to-average power ratio distribution along fiber in the optical OFDM transmissions,” in Proc. of Euro. Conf. Optical Commun., paper We.10.P1.69, (2011).

**33. **S. Barbarossa, M. Pompili, and G. B. Giannakis, “Channel-independent synchronization of orthogonal frequency-division multiple access systems,” IEEE J. Select. Areas Commun. **20** (2), 474–486 (2002) [CrossRef] .

**34. **ITU-T Recommendation G.975.1 2004, Appendix I.9.

**35. **J. G. Proakis and M. Salehi, *Digital Communications*, 5th ed., (McGraw-Hill International Editions, 2008).

**36. **Y. Yoshida, K. Hayashi, H. Sakai, and W. Bocquet, “Analysis and compensation of transmitter IQ imbalances in OFDMA and SC-FDMA systems,” IEEE Trans. Signal Process. **57** (8), 3119–3129 (2009) [CrossRef] .