1. Introduction

To satisfy the tremendous increase in capacity demand, 4 × 10-Gb/s TWDM-PON has been accepted as a primary solution to NG-PON2 standard following the commercial deployment of XG-PON and 10G-EPON [1, 2]. However, the 40-Gb/s capacity would not provide a long-term satisfaction for the desire of users and enterprises, so 100G optical access networks are widely investigated taking advantages of wavelength stacking techniques and advanced modulation formats such as NRZ with optical frequency equalization [1], duo-binary [3], pulse amplitude modulations (PAM) [4, 5], and direct detection optical orthogonal frequency division multiplexing (DDO-OFDM) [6]. Towards 100G beyond, more wavelengths and spatial channels can be multiplexed to further enhance the access capacity and scalability using multicore fibers (MCFs) or few mode fibers (FMFs) [7–10].

Aiming at settling dispersion induced power fading and stringent power budget problems in intensity modulation and direct detection (IM/DD) based high-speed long-reach access systems, coherent access network using digital signal processing (DSP) to compensate the linear impairments has been proposed [11, 12]. Coherent access can enhance the spectral efficiency and improve the receiver sensitivity, however, it brings about higher cost and complexity to optical network units (ONUs). Fortunately, self-homodyne coherent detection (SHCD) system, in which local oscillator (LO) is originated from the laser source at transmitter, provides a promising alternative with improved cost efficiency and simplified receiver-side DSP, because the tunable LO is not necessary and the laser linewidth requirement can be relaxed in SHCD system. Moreover, carrier frequency offset (CFO) and phase noise (PN) compensation algorithms can also be eliminated from the DSP due to the homology between the signal and the LO. Therefore, compared with traditional coherent detections, SHCD can greatly reduce not only the signal processing complexity but also latency, which makes it suitable for the emerging optical access network supporting 5G wireless service featured with high capacity, low latency, and low cost and power per bit. There are three main approaches to implement SHCD. One is to use two orthogonal polarizations to send signal and LO, respectively, which sacrifices half of the spectral efficiency [13]. Optical carrier extraction techniques can also be used to enable SHCD by separating the LO from the signal, but it puts stringent requirement on the narrow-band filter and also suffers from performance degradation due to the noisy LO [14]. Recent works on space division multiplexing (SDM) transmission systems have opened the possibility of employing SHCD with the signal and LO delivered via separated spatial channels either using FMF [15] or MCF [16]. MCF has advantages over FMF in aspect of channel counts, system stability and crosstalk between different spatial channels, thus can provide higher capacity and better performance. Although, in [16], the authors demonstrated an SHCD scheme in combination with 10.1-km 19-core fiber and free-space coupling systems, the transmission performance of 16 × 18 × 10-Gb/s QPSK signals is sensitive to the inter-core crosstalk because of the smaller core pitch in closely arranged 19-core fiber. Besides, the need of thermal isolation to ensure minimal variations of coupling efficiency and the free space optics based coupling devices with large size and weight may hinder their practical implementation. Most importantly, the data rate per wavelength is very limited, which not only makes the system susceptible to phase noise but also increases the average cost per bit of the coherent system. The expensive tunable external cavity laser (ECL) used in [16] is another obstacle before SDM enabled SHCD applied into cost-sensitive optical access networks. On the other hand, as coherent optical OFDM (CO-OFDM) is susceptible to CFO and PN, it is interesting to investigate the high baud rate 16QAM based CO-OFDM transmission performance in SDM enabled SHCD system thanks to the cancellation of frequency offset and phase noise.

In this paper, we propose a cost-reduced large-capacity access network employing SHCD with low-crosstalk MCF and compact fiber-based fan-in and fan-out devices, previously used in [7, 8]. In order to lower the bandwidth requirements in high-speed systems and improve the spectral efficiency, advanced quadrature amplitude modulation (QAM) formats are usually preferred. Although single carrier modulated QPSK, 16QAM or PAM4 signals are mature or cost-efficient in signal generation and reasonably more preferred for industry, compared with single carrier modulated systems, multi carrier modulation techniques show superior performances in terms of capacity flexibility, link adaptation capability using power and bit loading, fiber dispersion tolerance and simplicity of one-tap equalization. Considering the bandwidth exploding and big-data era we are living in, the data rate requirement and information infrastructure upgradation in access network segment is very common and necessary, which indicates that system scalability and upgradation compatibility inherently brought by multicarrier modulated schemes are meaningful to next generation optical access network. So, in our proposed schemes, polarization division multiplexed coherent optical orthogonal frequency division multiplexing (PDM-CO-OFDM) with 16QAM and direct detection optical filter bank multi carrier (DDO-FBMC) with 32QAM are used in the downstream (DS) and upstream (US) transmissions, respectively. For DS, information carrying signals are transmitted in the outer cores while the LO lights are delivered via the remained inner core of the MCF so that SHCD can be implemented in the ONU side. For the US, cost-effective IMDD scheme is enabled by spectrally efficient FBMC modulation format using low-bandwidth components. The proposed SHCD based access network is inherently compatible with other multiplexing methods like wavelength division multiplexing (WDM) and space division multiplexing (SDM), which can effectively increase the capacity. As a proof of concept, we experimentally demonstrated a $4\times 6\times 200$-Gb/s DS transmission over 37-km 7-core fiber without using CFO and PN compensation and $4\times 1\times 30$-Gb/s US transmission only using ~7 GHz-bandwidth receivers.

2. Architecture of the proposed WDM-SDM access network

The proposed hybrid wavelength division multiplexing and space division multiplexing (WDM-SDM) optical access network architecture utilizing MCF and SHCD is illustrated in Fig. 1. In the OLT block, m wavelengths are utilized as the laser source. For each wavelength in one subset OLT, it is power split by N + 1, one portion of which is used as the LO and the others are for signal modulation. In this way, this configuration can support N × m subscribers only employing m lasers in the OLT side which may significantly cut down the expense compared with the same situation in coherent WDM-PON. To enhance the capacity with affordable cost and complexity, downstream signal is suggested to be modulated with PDM-16QAM-OFDM, which is spectral efficient and bandwidth flexible with strong immunity to dispersion. After modulation, the N × m branches from ${\lambda }_1$ to ${\lambda }_m$are multiplexed respectively by N m-wavelength multiplexing devices like array waveguide gratings (AWGs). Afterwards, N sets of multi-wavelength signals are amplified by N erbium-doped fiber amplifiers (EDFAs). After going through the circulators, the N sets of signals are injected to the fan-in device which connects the single mode fibers (SMFs) and MCF. Meanwhile, the LOs are input to the inner core of the MCF in the same manner. After MCF transmission, the downstream signals together with LOs are output to N + 1 independent SMFs via the fan-out device. Subsequently, signals from each core are wavelength de-multiplexed and then coherently detected with the corresponding LO light at the ONU side. As in SHCD scheme the expensive tunable narrow-linewidth LO lasers and complicated CFO and PN compensation algorithms are no more required in the ONUs, the average cost per bit and system complexity can be much reduced considering the high data rate per user can enjoy. Moreover, with the development and maturing of optical amplification technologies in MCF, such as multicore erbium doped fiber amplifier (MC-EDFA) using cladding pumping [17] and multicore Raman amplifier using pump sharing [18], the cost of the total access system can be further reduced. Considering most of the asymmetric access scenarios, to support lower-speed upstream transmission, IMDD schemes with spectrally efficient modulation formats using low-bandwidth transponders are preferred. As a promising candidate for the upcoming 5G, FBMC can provide higher spectral efficiency due to its capability of operation without cyclic prefix (CP) and higher resilience to synchronization errors compared with its multicarrier counterpart of OFDM [19]. Constrained by the affordable power budget in US and signal to noise ratio (SNR) in IMDD system, 32-QAM can be a feasible choice for the US. To solve the wavelength sensitive problems in our proposed WDM-SDM access network, typical colorless ONU technologies can be deployed as many prior arts presented by either using tunable lasers [20] or reflective semiconductor optical amplifier (RSOA) based on external laser seeding [21–23]. To avoid the Rayleigh backscattering noise in the US, wavelengths for DS and US can be slightly offset in the spectrum domain.

 

Fig. 1 Schematic of proposed WDM-SDM access network based on MCF and SHCD. (CW: continuous wave, PS: power splitter, Rx: receiver, Mod: modulator, Mux: multiplexer, Demux: de-multiplexer, EDFA: erbium doped fiber amplifier, OC: optical circulator, MCF: multicore fiber, Tx: transmitter, PDM: polarization multiplexing, CO-Rx: coherent optical receiver, OLT: optical line terminal, ODN: optical distribution network, ONU: optical network unit.)

Download Full Size | PDF

3. Experimental setup and results

To verify the feasibility of our proposed scheme, we conduct bidirectional transmission experiments using the setup shown in Fig. 2. As shown in insets Fig. 2(a) and Fig. 2(b), we utilize 37-km low crosstalk 7-core fiber and compact low-loss fan-in/fan-out devices to implement the access network architecture. The design parameters of the used MCF is nearly the same as our previously published work in [7]. With improved operation platform and optimized manufacturing process, the performances of our fabricated 7-core fiber and fan-in/fan-out devices have been improved a lot. Table 1 shows the details of optical parameters comparison between the MCF and single mode fiber (SMF) fabricated by YOFC. As we can see, the properties of our MCF are comparable with that of the SMF. Considering the benefits such as low loss, loss crosstalk and numbers of parallel spatial channels, MCF is a promising physical media for next-generation high-speed wired and wireless converged optical access network supporting heterogeneous services. The insertion loss of a pair of fan-in/fan-out is reduced to about 2 dB per core. The end-to-end inter-core crosstalk and the total loss of the transmission link including 37-km MCFs and a pair of fan-in/fan-out devices can be as low as −60 dB and 10 dB, respectively, which are well enough for access network transmission.

 

Fig. 2 Experimental setup for WDM-SDM access network using SHCD and FBMC. (a) Cross section view of the fabricated seven-core fiber, (b) Picture of the overall fan-in device with single mode fiber pigtails. (Rx: receiver, Tx: transmitter, VOA: variable optical attenuator, PRBS: pseudo random binary sequence, PC: polarization controller, FBMC: filter bank multi carrier, AWG: arbitrary waveform generator, SMF: single mode fiber, PBS: polarization beam splitter, DSO: digital storage oscilloscope, DSP: digital signal processing, DF: decorrelation fiber)

Download Full Size | PDF

Table 1. Main optical parameters comparison between MCF and SMF

View Table

For SHCD DS transmission, the inner core (core 1) is dedicated for LO delivery and the other 6 cores are for signal transmission. The PDM IQ modulator and PDM coherent receiver are commercial components typically used in 100G PDM-QPSK systems. The sampling rate for the arbitrary waveform generator (AWG: Keysight, M8195A) and digital sampled oscilloscope (DSO: LeCory, LabMaster 10-36Zi-A) are fixed at 25-GS/s and 80-GS/s, respectively. The baseband PDM-16QAM-OFDM signal with 120 subcarriers is generated in Matlab using 128-point inverse fast Fourier transform (IFFT) and 10% CP, and the resultant gross data rate per wavelength is about 200-Gb/s. Constellation recovery and BER counting are implemented offline after OFDM demodulation using conventional PDM-CO-OFDM DSP algorithms [24]. Constant amplitude zero auto-correlation (CAZAC) precoding [25] is used in this experiment to improve the CO-OFDM transmission performance. As the bias voltage of the TIA for our coherent receiver in our lab is not optimized, the sensitivity of the coherent receiver is poor especially when the LO is weak. In order to maintain a good receiver sensitivity and ensure a relatively large dynamic range of received optical power, we used two EDFAs before the coherent receiver for the DS transmission in our experiment. For practical implementation, the EDFAs for LO and DS signals pre-amplification is not mandatory in the ONU side if high-sensitivity receivers are used.

For IMDD US transmission, as limited by the experimental condition, we only send 4 channels of 32-QAM-FBMC signals through core 2. The baseband 32-QAM-FBMC signal generation and demodulation are the same as in [26]. The procedure to generate real-valued FBMC signal is very similar to the generation of DDO-OFDM signals as we did in [27]. The only difference is the prototype filter inserted after IFFT. The prototype filter we used is poly phase network (PPN) filter and more details can be found in [28]. At the receiver, the channel estimation in FBMC is also worthy of being emphasized. Two typical features of the FBMC modulation are the mitigation of a cyclic prefix (CP) and the fact that the orthogonality property only holds in the real field. Consequently, the classical training symbol based channel estimation methods used in OFDM cannot be directly applied to FBMC. So in this work, specially designed training symbols [29] are used to avoid the crosstalk and improve the channel estimation performance. The signal processing at the transmitter side consists of serial-to-parallel (S/P) conversion, 32-QAM symbol mapping, CAZAC precoding, 128-point IFFT, PPN filtering, parallel-to-serial (P/S) conversion and digital-to-analog conversion (DAC). The signal processing at the receiver side includes analog-to-digital conversion (ADC), frame synchronization, serial-to-parallel (S/P) conversion, PPN filtering, 128-point FFT, channel estimation, CAZAC equalization and 32-QAM symbol decoding. Similar to DDO-OFDM signal generation, Hermitian symmetry should be applied to the PPN-FFT process in order to acquire real valued FBMC symbols, so only 63 subcarriers can be used to convey effective data. In this experiment, to be consistent with CO-OFDM based downstream transmission, the FBMC frame length is 139, 11 of which are training symbols used for frame synchronization and channel estimation. Sampling rate for AWG and DSO in US transmission are 12-GS/s and 40-GS/s, respectively and the aggregate US data rate is $4\times 1\times 30$ -Gb/s. Considering the fact that the photodetector (PD: Discovery Semiconductors, DSC-R402APD) we used in US only has a 3-dB bandwidth less than 7-GHz, the spectral efficiency of the system is higher than 4 bit/s/Hz. Note that, as is often the case, to avoid performance penalty, a sampling rate of 2 samples per symbol is needed in coherent optical communication systems. The sampling rate of the DSO should be 50 GS/s and 24 GS/s for DS and US, respectively. Due to the discrete tuning range of the DSO we used, sampling rates of 80 GS/s and 40 GS/s are used in our experiment. One more thing must be pointed out is that the bidirectional transmission is realized by sending dummy signals in DS and only demodulating the US signals when we are evaluating the US transmission performance, and vice versa. We should also note that 6 tributaries of downstream signals are time delayed by different lengths of decorrelation fibers for spatial channel decorrelation.

Firstly, we fix the power of LO at 13 dBm and compare the BER performances between traditional intradyne detection and the proposed SHCD scheme in a single wavelength MCF transmission system using ~100-kHz ECL. From the results shown in Fig. 3(a), we find that compared with SHCD, intradyne coherent detection can slightly improve the BER performance in the region of high received optical power (ROP), where the performance of SHCD is degraded by the slightly deteriorated optical signal to noise ratio (OSNR) of the LO during transmission and amplification. However, in the lower ROP range, intradyne detection will worsen the BER, which can be accounted for the penalty from the inaccuracy of the CFO and PN compensation under lower OSNR condition. As depicted in Fig. 3(b), we also evaluate the influence of CFO and PN compensation in SHCD. The CFO and PN compensation have no improvements in the ~100-kHz ECL based SHCD system because the LO and signal are derived from the same laser source and inherently coherent with identical phase evolution. Subsequently, we measure the transmission performances of SHCD without CFO and PN compensation for all the 6 cores using ~100-kHz ECL with launching power of 5 dBm and the results are summarized in Fig. 3(c). The receiver sensitivity to reach the BER limit (BER = 2.4e⁻²) of the soft-decision forward error correction (SD-FEC) coding [30] with 20% overhead is about −22 dBm, and for different fiber cores, it shows very similar BER performance, which proves the uniformity of our fabricated MCF. Although SD-FEC scheme is typically used for high-capacity core network currently, there is a trend that this technology is also possible to penetrate into access network in the future thanks to the increasing capabilities of application specific integrated circuits (ASICs) and development in digital signal processing (DSP) research area, just as coherent detection and advanced modulation formats did.

 

Fig. 3 (a) BER performance comparison between intradyne and SHCD, (b) Influence of CFO and PN compensation in SHCD, (c) BER performances for all cores.(w/:with, w/o: without, CFO: carrier frequency offset compensation, PN: phase noise compensation, SD-FEC: soft decision forward error correction coding)

Download Full Size | PDF

In order to demonstrate the WDM compatibility, four ECLs with channel spacing of 100 GHz (CH1:1550.116 nm, CH2:1550.916 nm, CH3:1551.716 nm and CH4:1552.516 nm) are combined by a WDM multiplexer and simultaneously modulated with 25-GBaud PDM-16QAM-OFDM signal. The spectra of signal and LO after MCF transmission are shown in Fig. 4(a). To demodulate each wavelength channel, the multiplexed signal and LO should be wavelength de-multiplexed after spatial de-multiplexing and amplification. Different from conventional polarization diverse intradyne coherent systems where the state of polarization (SOP) of the locally-generated LO can be easily aligned to the optical axis of the polarization beam splitter (PBS) of the optical hybrid, for SHCD systems, the SOP of the LO after transmission is unknown and time-varying [31]. A polarization controller (PC) is used on the LO tributary in front of the PDM coherent receiver in order to carefully align the polarization direction and avoid LO nulling in one of polarizations, in which case the demodulation of PDM signals cannot work correctly. As the polarization state changes slowly in fiber transmission systems, for practical applications, a commercial automatic polarization controller/tracker module can be used instead. As demonstrated in many previously published works [32–34], active polarization tracking and SOP control can be achieved using fiber squeezers and integrated optical components based on feedback from a PC, Lithium-Niobate polarization transformers and endless polarization tracking with additional phase control. Although these schemes have not been employed specifically with SHCD systems, it is envisaged that these extensively studied polarization tracking techniques can be easily applied to our SHCD systems. As all the outer cores in MCF have similar performance, for simplicity, we only plot the result of WDM signal transmission performance in core 7. The WDM transmission performances and recovered constellation diagrams on the orthogonal polarizations of CH3 are shown in Fig. 5(a). The result indicates that the downstream capacity can be increased to 4.8 Tb/s ($4\text{λ}\times 6\text{core}\times 200$Gb/s) by wavelength stacking technique and tiny implementation penalty is observed compared to single wavelength case.

 

Fig. 4 (a) Spectra of 4-channel PDM-16QAM-OFDM signals and LO lights in downstream transmission, (b) Spectra of 4-channel 32QAM-FBMC signals before and after modulation in upstream transmission.

Download Full Size | PDF

 

Fig. 5 (a) BER performance of WDM transmission in core 7 and the constellation diagrams at ROP = −15 dBm for channel 3 are inserted as insets, (b) Tolerance to laser linewidth in single wavelength SDM based SHCD system, and the insets are the recovered constellation diagrams at ROP = −17 dBm on X-polarization for 100-kHz and 10-MHz lasers. (w/: with, PN: phase noise compensation)

Download Full Size | PDF

After that, we investigate the tolerance to laser linewidth in single wavelength SDM based SHCD DS transmission. In our experiment, the LO path contains only one EDFA while the signal path has two. Typically, the erbium doped fiber in a commercial EDFA is about 10 m in length, so the coarse length difference between the LO and signal path is about 10 m. To ensure that the time offset between signal and LO path is within the coherence interval of 10-MHz linewidth DFB laser (~100 ns) and enable coherent demodulation without CFO and PN estimation, 15-m SMFs in the LO path is used as optical delay line to compensate the time delay. In practical SHCD systems, the time delay between the LO and signal can be estimated and pre-compensated with higher accuracy by sending a probe signal to measure the lengths of the two optical paths [16]. As for the path equalization, it is only necessary when large linewidth DFB lasers are used. If ~100 kHz linewidth ECLs are used as traditional coherent systems, the tolerance of optical path offset between LO and DS signals can be as large as 2000 m, which means the possibility of implementation without optical path compensation. As can be seen from Fig. 5(b), without CFO and PN compensation, $6\times 200$-Gb/s PDM-16QAM signal can be successfully transmitted along 37-km MCF using 10-MHz linewidth DFB laser with less than 2dB power penalty compared with 100-kHz ECL. It is also observed that with the help of PN compensation in the DSP, the transmission performance of the 10-MHz linewidth DFB laser based SHCD system can be slightly improved, especially in the large ROP range. This proves the feasibility of using cost-efficient DFB lasers with large linewidth in high order modulation format based large-capacity coherent access networks. Due to the high coherence between the signal and LO in SHCD system, strong interference induced unstable captured waveforms are observed during the experiment when the DS signal and LO are directly combined in the coherent receiver without fiber transmission, and we did not measure the transmission performance for optical back to back (OB2B) case. Moreover, considering the dispersion tolerant CO-OFDM modulation format we used in our SHCD based DS transmission, the power penalty after 37 km MCF transmission is not a critical issue.

Finally, we investigate the upstream transmission performance using 4 100-GHz spaced wavelengths multiplexed channels, whose spectra before and after FBMC modulation are shown in Fig. 4(b). As the wavelengths used in US and DS are different, Rayleigh backscattering noise in bidirectional transmission is avoided. In order to alleviate the signal to noise ratio (SNR) fading effect in the high frequency range, as is typically observed in bandwidth limited IMDD systems, we apply CAZAC precoding and equalization to the DDO-FBMC in the same way as [26, 27]. As expected, thanks to the linear precoding the estimated SNR across the whole bandwidth keeps almost flat both in optical back to back (OB2B) case and after MCF transmission regardless of the received optical power and wavelength channels, although the 3-dB bandwidth of the receiver is less than 7 GHz. The received constellation diagrams for different wavelength channel and received optical power are also depicted in Fig. 6(a). The BER performances for all the 4 wavelength channels are summarized in Fig. 6(b). Compared with OB2B, to reach the SD-FEC BER limit, about 1 dB power penalty is observed after 37 km MCF transmission. The receiver sensitivity is about −17 dBm for all the 4 channels of 30-Gb/s 32QAM–FBMC signals. The results unveil the potential of using high order spectrally efficient modulation formats to improve the system capacity with affordable complexity, especially in cost sensitive access networks.

 

Fig. 6 Transmission performance for upstream in both OB2B case and after 37-km MCF transmission. (a) Estimated SNR for different ROPs and wavelength channels across the entire signal bandwidth, (b) BER performance of all the four wavelength channels at variable received optical powers.

Download Full Size | PDF

We should point out that the data rate reported in this work is the aggregate data rate including the FEC overhead, training symbols, cyclic prefix (CP) and zero padded subcarriers. So if we calculate the net data rate, 20% FEC overhead should be firstly excluded, and then the training symbols, CP, and zero padded subcarriers. Taking all kinds of overheads into consideration, the net data rate per wavelength per core for DS is 125.4 Gb/s (80%$\times$120/128$\times$128/141$\times$128/139$\times$25$\times$2$\times$4), while the data rate per wavelength per core for US is 21.8 Gb/s (80%$\times$63/128$\times$128/139$\times$12$\times$5).

4. Conclusions

We proposed and experimentally demonstrated an ultra-high capacity WDM-SDM optical access network with 4.8-Tb/s downstream and 120-Gb/s upstream data rate over 37-km 7-core fiber utilizing SHCD OFDM and IM/DD FBMC, respectively. Compared with traditional intradyne systems, large capacity DS transmission is realized without expensive tunable narrow-linewidth LO lasers. The results also show that both cancellation of CFO and PN in MCF based SHCD system relaxes the requirement for narrow linewidth lasers and can greatly simplify the receiver-side DSP potentially enabling cost and power savings. The DDO-32-QAM-FBMC based upstream proves the feasibility of using high order multicarrier modulation formats in cost sensitive access scenario to realize high spectral efficiency even using bandwidth limited components. As SHCD improves the downstream receiver sensitivity and the spectrally efficient FBMC upstream ensures excellent dispersion resistance, in principle, our proposed SHCD and FBMC based WDM-SDM scheme can be easily extended to long-reach large-capacity access networks and metro networks with improved power budget using larger launch power. To support heterogeneous services converged optical access network including 5G and internet of things (IOT), MCF can provide an additional spatial dimension to realize high capacity with great flexibility. Besides, for practical applications, to further cut down the cost in upstream, cost-efficient approaches can be adopted to realize FBMC modulation by using direct modulation lasers (DML).

With the development of low crosstalk MCF and compact low loss fan-in and fan-out devices, more spatial channels can be feasible in one pipeline thus it emerges as an ideal platform for future data-driven multi-service featured communications. Further, taking advantages of the multicore fiber amplifiers such as multicore EDFA and Raman amplifiers, the system complexity can be much simplified with lower cost. Ultimately, dense spatial division multiplexing technologies using multicore few mode fibers are expected to penetrate into access network to enhance the capacity by orders of magnitude supporting more and more end subscribers with lower cost per user per unit data rate.