1. Introduction

Due to the added cost of optical amplifiers, their use in commercial optical access deployments has been limited to systems where they are heavily shared among a large number of subscribers. For example, high-power erbium-doped fiber amplifiers (EDFAs) are routinely used as boosters in hybrid-fiber coax (HFC) systems [1] and RF-video overlays of passive optical networks (PONs) [2]. In both cases, RF subcarrier-modulated optical signals in the 1550-nm band, requiring high electrical signal-to-noise ratio and therefore high received optical power, are broadcast to a thousand or more subscribers, offsetting the cost of the amplified transmitter.

Commercially available time-division multiplexed (TDM) PONs operate at peak rates up to 10 Gb/s [3], use robust baseband on-off keyed modulation, and have far fewer subscribers sharing the transceiver at the optical line terminal (OLT). Rather than employing optical amplifiers, these systems meet loss budget requirements with directly modulated transmitter lasers and (in some cases) sensitive avalanche photodiode (APD) receivers. Thus, research on amplified TDM PONs has focused on long-reach, high-split architectures developed in anticipation of future standards [4], or in recent years as a long-reach option to the gigabit-capable passive optical networks (GPON) standard [5, 6]. These approaches, although technically feasible, have not yet gained acceptance in the market. One likely explanation is the reluctance of telecom carriers to add powered elements to the traditionally passive PON optical distribution network (ODN). With this constraint in mind, remotely pumped EDFAs [7, 8] and distributed Raman amplification [9] have been proposed as solutions having the desirable aspects of inline amplifiers without compromising the passive ODN. The distributed Raman approach has been incorporated into the GPON standard [10] and has been the basis of a system demonstration showing coexistence of extended GPON and 10G-PON [11].

In stark contrast to commercial TDM PONs, the time and wavelength-division multiplexed (TWDM) PON recently defined in ITU-T’s NG-PON2 standard [12] has symmetric rate versions with as many as eight 10-Gb/s wavelengths in each direction, and includes maximum optical path losses of 29 dB, 31 dB, 33 dB, and 35 dB, for ODN classes N1, N2, E1, and E2, respectively. To meet the most aggressive of these optical path loss requirements, optical amplifiers may be incorporated into the OLTs as downstream boosters and upstream pre-amplifiers [13]. Locating both upstream and downstream amplification at the OLT is considered preferable to adding one or more amplifiers to each user’s optical network unit (ONU), since the costs of OLT optics are shared among all users. This opens the door for distributed Raman solutions, for which a single pump wavelength can efficiently amplify a band of WDM channels. As compared to lumped amplification at the OLT, distributed amplification reduces the peak signal power of downstream signals, thereby limiting nonlinear impairments, and improves the effective noise figure for upstream amplification [14]. Furthermore, residual pump light can be harvested to provide electrical powering for active functions such as monitoring [15] and switching [16].

Here we demonstrate a symmetric-rate 1:128 split TWDM PON with OLT-based bi-directional distributed Raman amplification and eight 10-Gb/s wavelength channels per direction. We measure bit-error-rate (BER) performance for three feeder lengths: 10 km, 21 km, and 42 km, and show that Raman gain scales with feeder length to offset feeder losses.

2. Experiment

The Raman-amplified TWDM PON architecture, shown schematically in Fig. 1, is based on the architecture defined by the NG-PON2 standard [12], with the addition of Raman pump lasers and pump multiplexers, that add Raman pump light to the feeder fiber at the OLT, and a pump demultiplexer to remove pump light before the splitter. Figure 2 is a more detailed representation of the experimental setup. The downstream and upstream wavelength channels used for the experiment conform to the operating bands defined in the standard and have 100-GHz channel spacing. The downstream band consists of eight L-band ITU grid wavelengths from λ_1D at 1597.19 nm to λ_8D at 1603.17 nm, and the upstream band consists of eight C-band ITU grid wavelengths from λ_1U at 1538.98 nm to λ_8U at 1544.53 nm. Although commercial systems built to the standard will most likely employ directly modulated distributed feedback (DFB) lasers or electro-absorption-modulated lasers (EMLs), we use fixed DFB laser transmitters and external Mach-Zehnder modulators (MZMs) for both the upstream and downstream transmitters. This approach ensures uniform chirp-free signals under test and allows us to investigate the change in BER performance associated with Raman amplification as compared to the best-case non-amplified performance. The eight downstream wavelengths are modulated with a 10.644-Gb/s non-return to zero signal in a common MZM. The 10.644-Gb/s rate, which is higher than the NG-PON2 standard rate of 9.95328 Gb/s, was chosen for compatibility with our commercial receivers. The pattern length of the pseudorandom bit stream is 2¹⁵-1. The channels are amplified with a two-stage L-band EDFA and decorrelated with a mid-stage 1:4 wavelength-selective switch (WSS), as shown in the inset to Fig. 2. The WSS directs wavelength pairs (λ_1D λ_5D), (λ_2D λ_6D), (λ_3D λ_7D), and (λ_4D λ_8D), to output ports 1, 2, 3, and 4, respectively. These are then connected with unequal fiber lengths to a 4:1 passive combiner, thereby decorrelating neighboring channels. The EDFA amplification offsets the losses of the 8:1 passive combiner, MZM, and WSS-based decorrelator. Launch powers are set to 3 dBm per channel entering the feeder fiber in agreement with the minimum mean launch power defined by the standard for 10-Gb/s channels. Thus, the transmitter configuration emulates the lowest-power, and therefore lowest-cost, transmitter option. Grating-stabilized DFB laser pumps produce pump light at 1455 nm and 1510 nm, which is added to the feeder fiber with dedicated pump multiplexers. The loss experienced by the signal channels traversing both these multiplexers is 0.7 dB, but can be reduced below 0.5 dB with a custom multiplexer as shown schematically in Fig. 1. These pump wavelengths were selected based on availability, and can be optimized. Several feeder fiber lengths were tested: 10 km, 21 km, and 42 km of depressed-water-peak standard single-mode fiber. The remote node consists of a pair of pump demultiplexers to strip the residual pump light from the feeder, followed by a 1:128 passive splitter. As with the pump multiplexers, the 1.0-dB measured loss through the pair of pump demultiplexers can be reduced with a single 3-port device to demultiplex both pump wavelengths. For the base architecture shown in Fig. 1, each splitter port is connected to an ONU that can tune to any pair of the eight upstream and downstream wavelength channels. For the purposes of our experiment (Fig. 2), we simultaneously modulate all eight upstream wavelengths at 10.644 Gb/s with a set-up identical to that used for the downstream transmitter including decorellation and EDFA amplification. Thus, all eight upstream wavelengths are launched into a single drop fiber with a per-channel power of 2 dBm after the diplexer, again in accordance with the minimum mean launch defined by the standard. For experimental convenience, both the OLT and ONU receivers include ITU-grid arrayed-waveguide grating (AWG) demultiplexers and commercial XFP-based APD receivers. A realistic commercial ONU would employ a tunable directly-modulated laser and tunable filter.

 

Fig. 1 Raman-amplified TWDM PON architecture.

Download Full Size | PDF

 

Fig. 2 Experimental setup.

Download Full Size | PDF

The Raman pump powers entering the feeder after the second pump coupler are set to 340 mW at 1455 nm and 220 mW at 1510 nm. Lower pump power is required for the 1510-nm pump since power is transferred from the 1455-nm pump to the 1510-nm pump via the Raman process. These values were chosen to provide sufficient gain for the 42.4-km system, while maintaining moderate total pump powers. For the 42.4-km case, the L-band (downstream) gain, measured with 1510-nm pumping alone, increases by more than 4 dB to 9.8 dB when the 1455-nm pump light is added. With both pumps present, the C-band (upstream) gain decreases by ~0.2 dB to 8.6 dB, due to 1455-nm pump depletion. Slightly higher gain was required for the downstream direction to compensate for higher loss experienced by the downstream signals including 1.9 dB higher loss in the downstream AWG demultiplexer as compared to the upstream AWG demultiplexer. With the same pump powers, the downstream and upstream gains for the 21-km system were 7.3 dB and 7.3 dB, respectively, while those for the 10-km system were 4.3 dB and 5.0 dB. Although the Raman gain decreases with decreasing feeder length, the magnitude of the decrease is less than the reduction in system loss associated with the length reduction. For example, decreasing the feeder length by 21-km reduces the total loss of the system by > 4 dB, but the Raman gain is only reduced by 2.5 dB downstream and 1.3 dB upstream. Thus designing the Raman gain for the longest system reach insures operation with additional margin for shorter reaches.

Table 1 shows residual pump power measured at the outputs of the pump demultiplexers at the end of the feeder for each feeder length. When increasing the feeder length from 10.0 km to 42.4 km, the residual pump power at 1455 nm decreases from 19.5 dBm to 9.0 dBm, while that at 1510 nm decreases from 22.0 dBm to 15.9 dBm. The total residual pump power exceeds 16 dBm for all three feeder lengths and is removed before the splitter to avoid the need for pump blocking filters at the ONUs. This residual pump light is wasted in our system; however as mentioned previously, residual pump light can be used to power active components collocated with the passive splitter [15–17]. As a follow-up to this work, we recently demonstrated an intelligent TWDM PON that uses Raman pump light to provide amplification for upstream PON signals while simultaneously powering an intelligent splitter module (ISM) and serving as the telemetry channel to the ISM [18].

Table 1. Residual pump powers for various feeder fiber lengths

View Table

3. System performance

Bit-error rate (BER) measurements were performed with all sixteen wavelengths operating for each feeder length. Three of the eight channels were measured in each band: the shortest and longest wavelength channels and one middle channel. At each distance, the variation in performance among the channels in each band was small (< 0.3 dB at 1 x 10⁻³ BER, the raw BER corresponding to a post-forward error correction (FEC) BER of 1 x 10⁻¹² for Reed-Solomon (248, 216) FEC as specified by the standard), so Figs. 3 and 4 show downstream and upstream BER data, respectively, for a middle channel only. Aside from a slight difference in absolute sensitivity between the two commercial XFP receivers, both the downstream and upstream data sets look very similar: at 1 x 10⁻³ BER there is negligible difference between 1) single-channel back-to-back receiver sensitivity, 2) back-to-back curves with all eight EDFA-amplified and decorrelated wavelengths working, 3) the fully loaded system (eight downstream wavelengths working, eight upstream wavelengths working, 1:128 split, and bi-directional distributed Raman amplification) with 10-km feeder, and 4) the fully loaded system with 21.2-km feeder. For both downstream and upstream there is a small penalty of less than 0.5 dB at 1 x 10⁻³ BER for the fully loaded 42-km system, resulting in worst-case sensitivities equal to or exceeding −30.6 dBm. Interestingly, the causes of these small penalties are quite different. Measured BER performance (not shown) for a single 10-Gb/s channel over the 42-km feeder overlaid the back-to-back curves, indicating that chromatic dispersion alone is not the cause of the power penalties for the 42-km fully loaded system. The cause of the upstream penalty is degradation in optical signal-to-noise ratio (OSNR). The upstream channels experience 22.5-dB loss traversing the 1:128 splitter before entering the Raman-pumped feeder fiber. At the OLT, the measured OSNRs are 31.4 dB, 29.0 dB, and 25.0 dB for the 10-km, 21-km, and 42-km systems, respectively. We used the transmitter EDFA to noise-load the 1542.17-nm channel in a back-to-back configuration, and confirmed that the upstream power penalty shown in Fig. 4 for the 42-km system is entirely due to OSNR degradation. The downstream channels have received OSNRs exceeding 36 dB, since the channel powers entering the feeder are relatively high, and therefore experience no power penalty due to OSNR degradation. Rather, the slight downstream power penalty is attributed to the interaction of cross-phase modulation and chromatic dispersion. The spectra of the Raman-amplified downstream channels are broadened by pump-signal cross-phase modulation, resulting in a measureable chromatic dispersion penalty for the 42-km system.

 

Fig. 3 Downstream bit-error rate (BER) performance for the 1599.75-nm channel.

Download Full Size | PDF

 

Fig. 4 Upstream bit-error rate (BER) performance for the 1542.17-nm channel.

Download Full Size | PDF

Although we did not directly test the upstream transmission with burst-mode signals, we did measure the upstream saturation performance of the distributed Raman amplification using CW signals over a broad range of input powers for a 42.4-km feeder fiber. Increasing the total upstream power launched into the feeder from −12.4 dBm (−21.4 dBm per channel) to 15.6 dBm (6.6 dBm per channel) resulted in 0.2 dB gain compression. Assuming a PON with a moderate 1 x 8 splitting ratio having 10-dB loss, this corresponds to per-channel transmitter powers > 16 dBm, 7 dB above the maximum mean launch power permitted by the standard. Thus is it reasonable to conclude that cross-gain modulation via the distributed Raman process should not limit burst-mode performance.

4. Conclusions

Trends in optical access systems toward longer reach, higher splitting ratio, and higher rates are driving solutions that incorporate optical amplification as both downstream boosters and upstream pre-amplifiers. Distributed Raman amplification, that heretofore has been exclusively deployed in metro and core systems, may be well-suited to future multiwavelength access networks due to its relative immunity to gain saturation-induced WDM crosstalk and superior effective noise figure.

We have demonstrated a Raman-amplified, symmetric-rate, 80 x 10-Gb/s TWDM PON conforming to the channel bands and minimum transmitter powers defined by the NG-PON2 standard. Using moderate Raman pump powers, the Raman gain scaled with feeder length such that an experimental system designed for 42-km reach and 1:128 split, had sufficient gain with additional margin at shorter reaches of 21 km and 10 km. BER performance varied only slightly with reach. A small penalty in both directions at the maximum reach of 42 km was attributed to OSNR degradation upstream and a combination of cross-phase modulation and chromatic dispersion downstream. The residual pump light arriving at the splitter exceeds 16 dBm for all three feeder lengths. This residual pump light can be used to power active subsystems, while maintaining a non-electrically powered outside plant.