1. Introduction

Rapid growth of internet traffic, driven by the proliferation of video services, continues to push broadband optical access networks for higher data rates and better quality of services. Currently GPON and EPON have been deployed worldwide for broadband services, providing aggregated bandwidth up to 2.5 Gb/s [1,2]. To satisfy the ever increasing bandwidth demands from end users, next generation 10G PONs (i.e. XG-PON and 10G EPON) with aggregated bandwidth of 10 Gb/s have been standardized and are ready for large scale deployment [3,4]. Beyond 10G PONs, future generation of optical access technologies, including WDM (Wavelength Division Multiplexed) PON, TWDM (Time and Wavelength Division Multiplexed) PON, OFDM (Orthogonal Frequency Division Multiplexed) PON and OCDM (Optical Code Division Multiplexed) PON, has been proposed and demonstrated with aggregated bandwidth at 40 Gb/s or higher [5–8]. Among these, TWDM PON has been selected by FSAN (Full Service Access Network) community as a primary broadband solution for future access networks [9]. Such TWDM PON systems can provide 40 Gb/s aggregated capacity with 1:64 splitting ratio and 40km reach [6,10], meeting the operators’ requirements for future broadband services [11]. A new set of standards on TWDM PON is expected to be released by ITU-T very soon [12], paving the way for future large scale deployment.

With multiple wavelengths and wavelength tunability, TWDM PON allows enhanced network functionalities unavailable in previous generations of pure TDM PONs. Incremental bandwidth upgrade and load balancing have been achieved in TWDM PON systems using fast tunable transmitters on OLT (optical line terminal) line card [13,14]. Energy efficient solutions in TWDM PONs were also demonstrated with dynamic wavelength routing [15]. In this paper, we propose and demonstrate a flexible TWDM PON architecture which allows pay-as-you-grow in capacity, load balancing among different ODNs (optical distribution networks), and power saving at OLT. With only passive components added in the ODNs, the proposed flexible TWDM PON supports pay-as-you-grow deployment of OLT transceiver modules and smooth upgrade of the aggregated capacity in each ODN. Load balancing for improved network performance can be achieved in a single ODN and among different ODNs with flexible lambda connections. Furthermore, selective OLT sleep for power saving is supported with narrow wavelength tuning (100 GHz) in OLT transceivers. Hence, our flexible TWDM PON system prototype exhibits enhanced functionalities and offers more flexibility compared to previously demonstrated system. Not only does it allow load balancing in a single ODN, but also provides bandwidth wherever is needed among multiple ODNs. In addition, this flexible TWDM PON achieves significant power saving at OLT side and provides improved resilience against OLT transceiver failures.

In addition to the proposed flexible TWDM PON architecture, we also developed, for the first time, pluggable OLT and ONU (optical network unit) optical transceiver modules for cost effective deployment of TWDM PONs. The OLT transceiver in enhance CFP (C Form Factor Pluggable) module consists of 4 × 10-Gb/s transmitters, 4 × 2.5-Gb/s burst-mode optical receivers, optical multiplexer/demultiplexer and integrated optical amplifiers. The low-cost ONU transceiver in SFP+ (Small Form Factor Pluggable) package includes a tunable laser and a tunable receiver. With these pluggable transceiver modules, a flexible TWDM PON system is demonstrated for error-free performance with more than 36dB power budget.

2. Flexible TWDM PON architecture

Figure 1 shows the architecture of the standard TWDM PON system, which uses both wavelength and time division multiple access. In such a system, multiple wavelengths coexist in the same ODN using wavelength division multiplexing, and each wavelength serves multiple ONUs with time division multiple access. As defined in ITU-T draft recommendation, 4~8 wavelengths on ITU grid in L band (1590-1610 nm), each running at 10 Gb/s data rate, are used for downstream transmission. Similarly, 4~8 wavelengths in C band (1520-1540 nm) are used for upstream transmission, each carrying 2.5 Gb/s or 10 Gb/s data [12]. An ONU in TWDM PON is equipped with a tunable transceiver, so it can selectively transmit/receive upstream/downstream data on a pair of upstream/downstream wavelengths. This avoids ONU inventory issue and supports load balancing within the TWDM PON. Essentially, TWDM PON is stacked XG-PONs with each XG-PON running on different wavelengths, and the GEM (GPON encapsulation method) frame used in GPON and XG-PON can be reused for TWDM PON. As TWDM PON system can coexist with GPON/EPON and 10G PONs, the evolution of optical access toward TWDM PON will be very smooth with little service interruption.

 

Fig. 1 Standard TWDM PON architecture.

Download Full Size | PDF

Based on the standard TWDM PON, we propose a flexible TWDM PON architecture as shown in Fig. 2. Compared to the conventional TWDM PON system in Fig. 1, a hybrid AWG (Arrayed Waveguide Grating)/splitter is inserted between the OLT and ODNs. The hybrid AWG/splitter consists of a 4 × 4 cyclic AWG and a 4 × 4 power splitter, whose inputs and outputs are connected to 100 GHz optical interleavers. Even though these additional components introduce extra loss, 20km reach and 1:64 splitting ratio can be achieved with a bidirectional optical amplifier in the central office, as demonstrated in the following experimental results. Inside the OLT transceiver module, 4 transmitters send 10 Gb/s downstream signals, respectively, on a wavelength set at${\lambda }_1^d,{\lambda }_2^d,{\lambda }_3^d$and${\lambda }_4^d$with 200 GHz spacing (superscript d denotes downstream). The transmitters can be thermally tuned by 100 GHz, and emit another wavelength set,${\lambda }_1^{d+},{\lambda }_2^{d+},{\lambda }_3^{d+}$and${\lambda }_4^{d+}$. A wavelength-independent combiner, composed of a polarization beam combiner and a 3-dB coupler, can be used to multiplex the 4 transmitter wavelengths. Because of the 100 GHz interleavers in the hybrid AWG/splitter, the wavelength set, ${\lambda }_1^d,{\lambda }_2^d,{\lambda }_3^d$and${\lambda }_4^d$, will go through the cyclic AWG with 200 GHz channel spacing, so that each wavelength will be distributed to different ODNs. As the same wavelength from different modules will go to different output ports of the cyclic AWG (e.g.${\lambda }_{1,\text{M1}}^d$ from transceiver module M1 goes to ODN1, ${\lambda }_{1,\text{M2}}^d$ from transceiver module M2 goes to ODN2, and so on), each ODN will get a unique wavelength set ${\lambda }_1^d,{\lambda }_2^d,{\lambda }_3^d$and${\lambda }_4^d$, with each wavelength coming from a different transceiver module. For example, in ODN1, ${\lambda }_1^d$comes from transceiver module M1, ${\lambda }_2^d$ from transceiver module M4, ${\lambda }_3^d$ from transceiver module M3 and ${\lambda }_4^d$ from transceiver module M2. On the other hand, the wavelength set, ${\lambda }_1^{d+},{\lambda }_2^{d+},{\lambda }_3^{d+}$and${\lambda }_4^{d+}$, will be directed to the power splitter by the interleavers, and all of these wavelengths will be distributed to all 4 ODNs through the 4 × 4 power splitter. Similarly, in the upstream transmissions, ONU wavelength can be tuned to two set of wavelengths, $\{{\lambda }_1^u,{\lambda }_2^u,{\lambda }_3^u,{\lambda }_4^u\}$ and $\{{\lambda }_1^{u+},{\lambda }_2^{u+},{\lambda }_3^{u+},{\lambda }_4^{u+}\}$, with superscript u denoting upstream.. Both sets of wavelengths have a channel spacing of 200 GHz and the second set of wavelengths is shifted by 100 GHz with respect to the first set. Thus, wavelength set,$\{{\lambda }_1^u,{\lambda }_2^u,{\lambda }_3^u,{\lambda }_4^u\}$, will go through the cyclic AWG and the other set, $\{{\lambda }_1^{u+},{\lambda }_2^{u+},{\lambda }_3^{u+},{\lambda }_4^{u+}\}$, will pass through the power splitter. At OLT receiver, the demux has a 200 GHz bandwidth, so both ${\lambda }^u$(a wavelength from the first set) and ${\lambda }^{u+}$ (a wavelength from the second set that is shifted by 100 GHz from ${\lambda }^u$) are sent to a single receiver. A narrowly tuned optical filter inside ROSA (receiver optical subassembly) can select either ${\lambda }^u$ or ${\lambda }^{u+}$. Note that tunable filter might not needed as will be demonstrated in section 2.2. The advantages of such an architecture include (a) pay-as-you-grow for incremental bandwidth upgrade, (b) load balancing and on-demand bandwidth provision, (c) power saving at OLT, and (d) resilience against OLT transceiver failures.

 

Fig. 2 Flexible TWDM PON architecture.

Download Full Size | PDF

2.1 Pay-as-you-grow deployment

TWDM PON is considered as the future generation of optical access following 10G PON systems. Hence, it must be backward compatible with legacy PON systems. Not only must TWDM PON coexist with legacy PON systems, but also smooth evolution from 10G PON toward TWDM PON is necessary without any service interruption. In addition to seamless upgrade, the proposed TWDM PON architecture also supports pay-as-you-grow deployment of 40G TWDM PONs, as illustrated in Fig. 3. Suppose there are four 10G PON systems initially deployed in the field, each serving a single ODN with 10 Gb/s capacity. As user bandwidth demands keep increasing, upgrading toward 40G TWDM PON is required at some point. A straightforward approach would be deploying four 40G TWDM PON on day one, one for each ODN. However, this would incur significant investment from the beginning and a sudden increase in bandwidth from 10 Gb/s to 40 Gb/s might not be necessary as user bandwidth demands increase gradually over time. Hence, a more practical approach is to increase the capacity gradually from 10 Gb/s to 20 Gb/s, then from 20 Gb/s to 30 Gb/s, and eventually to the full capacity of 40G TWDM PON. As shown in Fig. 3(a), a 40 Gb/s OLT transceiver module (same as shown in Fig. 2) for TWDM PON is activated when bandwidth upgraded is necessary. Four wavelengths from this module (OLT module M1), ${\lambda }_{1,\text{M1}}^d,{\lambda }_{2,\text{M1}}^d,{\lambda }_{3,\text{M1}}^d$and${\lambda }_{4,\text{M1}}^d,$each with 10 Gb/s data rate, are separated by the 4 × 4 cyclic AWG and sent to 4 different ODNs. More specifically, ${\lambda }_{1,\text{M1}}^d$ is sent to ODN1, ${\lambda }_{2,\text{M1}}^d$ to ODN2, ${\lambda }_{3,\text{M1}}^d$ to ODN3, and ${\lambda }_{4,\text{M1}}^d$ to ODN4. Combined with the original 10G PON system, the total capacity for each ODN is now 20Gb/s. Later, as the user bandwidth increases further, a second TWDM PON transceiver module (M2) can be activated, as shown in Fig. 3(b). Again, four wavelengths from this module are separated by the cyclic AWG, each serving a different ODN. That is, ${\lambda }_{1,\text{M2}}^d$from module M2 is sent to ODN2, ${\lambda }_{2,\text{M2}}^d$to ODN3, ${\lambda }_{3,\text{M2}}^d$to ODN4, and${\lambda }_{4,\text{M2}}^d$to ODN1. The total capacity is then increased to 30Gb/s for each ODN. Gradually a third 40G TWDM PON transceiver module (M3) can be added to meet the increasing bandwidth demands, as shown in Fig. 3(c). Eventually, when a fourth transceiver module (M4) is activated, a total capacity of 50Gb/s in each ODN can be achieved with a 40G TWDM PON plus the original 10G PON (Fig. 3(d)). With such a modular approach, gradual bandwidth upgrade is achieved with low initial CAPEX investment for TWDM PON deployment. Note that the 4 × 4 splitter and the interleavers in the hybrid AWG/splitter can be removed without affecting the pay-as-you-grow deployment, but these components are needed for load balancing and power saving when the TWDM PONs are fully deployed.

 

Fig. 3 Modular approach for pay-as-you-grow deployment of TWDM PON. (a) One TWDM PON OLT transceiver module (M1) is added to increase the capacity in each ODN to 20Gb/s. (b) A second TWDM PON OLT transceiver module (M2) is activated for 30Gb/s capacity in each ODN. (c) A third TWDM PON OLT transceiver module (M3) is added for 40Gb/s capacity in each ODN. (d) A fourth TWDM PON OLT transceiver module (M4) is activated and the total capacity in each ODN reaches 50Gb/s. WDM filters in the figures separate the 10G PON wavelengths from TWDM PON wavelengths.

Download Full Size | PDF

2.2 Load balancing and OLT transceiver protection

In standard TWDM PONs, each ONU is equipped with a tunable transceiver, so load balancing is possible within a single ODN (i.e. a single PON). If the traffic load in a specific downstream/upstream wavelength pair is too heavy, then a certain number of ONUs using this wavelength pair could switch to another pair with less traffic load. In the flexible TWDM PON test bed demonstrated in section 2.4, each ONU uses a fast tunable DBR laser as its transmitter. Assume that 125μs upstream frame is adopted in TWDM PON (same as in GPON and XG-PON), burst-by-burst load balancing within a single TWDM PON could be achieved using tunable DBR lasers.

In addition to load balancing within a single ODN, the proposed flexible TWDM PON also allows load balancing among different ODNs, as shown in Fig. 4. Initially, a set of four wavelengths with 200 GHz spacing, ${\lambda }_{1,\text{M1}}^d$, ${\lambda }_{2,\text{M4}}^d$, ${\lambda }_{3,\text{M3}}^d$ and${\lambda }_{4,\text{M2}}^d$, (subscripts M1, M2, M3 and M4 denote which module the wavelength is from), serves ODN1, and another set of four wavelengths, ${\lambda }_{1,\text{M2}}^d$, ${\lambda }_{2,\text{M1}}^d$, ${\lambda }_{3,\text{M4}}^d$ and ${\lambda }_{4,\text{M3}}^d$, serve ODN2, and so on (as shown in Fig. 2). Suppose for a specific period of time, the traffic load in ODN1 is low, and a single wavelength with 10 Gb/s capacity is enough to serve the users in ODN1. So we could use a single wavelength, ${\lambda }_{1,\text{M1}}^d$, to serve ODN1 and leave other wavelengths (${\lambda }_{2,\text{M4}}^d$, ${\lambda }_{3,\text{M3}}^d$ and${\lambda }_{4,\text{M2}}^d$) idle. On the other hand, if ODN2 has very heavy traffic, a set of four wavelengths (${\lambda }_{1,\text{M2}}^d$, ${\lambda }_{2,\text{M1}}^d$, ${\lambda }_{3,\text{M4}}^d$ and ${\lambda }_{4,\text{M3}}^d$for a total capacity of 40Gb/s) is not enough to support the user bandwidth demands. In this case, the wavelengths ${\lambda }_{2,\text{M4}}^d$, ${\lambda }_{3,\text{M3}}^d$ and${\lambda }_{4,\text{M2}}^d$, which normally serve ODN1 but become idle because of the low traffic load in ODN1, will be shifted by 100 GHz to ${\lambda }_{2,\text{M4}}^{d+}$, ${\lambda }_{3,\text{M3}}^{d+}$ and${\lambda }_{4,\text{M2}}^{d+}$, by thermal tuning. Now these downstream wavelengths will go through the power splitter instead of the cyclic AWG, and thus reach ODN2; in other words, these wavelengths can now be used to serve users in ODN2. Similarly, some of the ONU transmitters in ODN2 can now be tuned to ${\lambda }_{2,\text{O2}}^{u+},{\lambda }_{3,\text{O2}}^{u+}$and${\lambda }_{4,\text{O2}}^{u+}$(subscript O2 denotes that the wavelength is from ONUs in ODN2). These upstream wavelengths will also go through the splitter instead of the AWG, and reaches the receivers, Rx2(M2), Rx3(M3) and Rx4(M4), which normally serve ODN1. Correspondingly, the tunable filter inside the RSOAs for Rx2(M2), Rx3(M3) and Rx4(M4) will be tuned to${\lambda }_2^{u+},{\lambda }_3^{u+}$and${\lambda }_4^{u+}.$ By doing so, three additional wavelength pairs, each with 10 Gb/s data rate, now serve ODN2. Thus, the total capacity in ODN2 increases to 70 Gb/s.

 

Fig. 4 Load balancing in the flexible TWDM PON. As the traffic load in ODN1 is low, TRx2, TRx3 and TRx4 in transceiver module M4, M3 and M2 respectively are tuned to${\lambda }_{2,\text{M}4}^{d+}$, ${\lambda }_{3,\text{M}3}^{d+}$ and${\lambda }_{4,\text{M}2}^{d+}$, so that these transceivers can serve ODN2 which has heavy traffic load.

Download Full Size | PDF

To achieve load balancing, tunable filters with 100GHz tuning range are used in all the receivers inside the OLT transceiver module, as discussed in the previous paragraph. For simpler receiver design, it is also possible to use fixed receiver at OLT. Furthermore, we could remove the 4 × 4 splitter and replace it with 4 fibers that bypass the cyclic AWG, as shown in Fig. 5. Consider the same case that ODN1 has only 10 Gb/s traffic load, while ODN2 requires 70Gb/s bandwidth. To achieve that, the wavelengths of all the transmitters in module M2 will be shifted by 100 GHz to ${\lambda }_{1,\text{M2}}^{d+},{\lambda }_{2,\text{M2}}^{d+},{\lambda }_{3,\text{M2}}^{d+}$and${\lambda }_{4,\text{M2}}^{d+},$ so that all these wavelengths will bypass the cyclic AWG and goes directly to ODN2. Similarly, the wavelength of the transmitter in TRx3 of the module M3 will be shifted by 100 GHz to ${\lambda }_{3,\text{M3}}^{d+}$, and the wavelength of the transmitter in TRx2 of module M4 will be shifted by 100 GHz to ${\lambda }_{2,\text{M4}}^{d+}$, so that the wavelength ${\lambda }_{3,\text{M3}}^{d+}$ goes to ODN3, and the wavelength ${\lambda }_{2,\text{M4}}^{d+}$ goes to ODN4. As the WDM multiplexers in the transceivers have 200 GHz channel spacing, the wavelengths remain in the same WDM channels after 100 GHz wavelength shift for all the transmitters in module M2, and the transmitters in TRx3 of module M3 and TRx2 of module M4. Alternatively, the WDM multiplexers can be replaced with wavelength-agnostic passive optical couplers, so that transmitter wavelength shift does not impact the WDM functionality. Note that the wavelength of the transmitter in TRx1 in module M2 can be kept unchanged as ${\lambda }_{1,\text{M2}}^d$or tuned to ${\lambda }_{1,\text{M2}}^{d+}$. With such a wavelength assignment shown in Fig. 5(a), there is only 1 wavelength serving ODN1, but 7 wavelengths are serving ODN2, 4 wavelengths serving ODN3, and 4 wavelengths serving ODN4. Therefore, the downstream capacity for ODN2 is increased to 70 Gb/s. For the upstream, ONU transmitters in ODN2 can be tuned to either ${\lambda }_{1,\text{O2}}^{u+}({\lambda }_{1,\text{O2}}^u),{\lambda }_{2,\text{O2}}^u,{\lambda }_{2,\text{O2}}^{u+},{\lambda }_{3,\text{O2}}^u,{\lambda }_{3,\text{O2}}^{u+},{\lambda }_{4,\text{O2}}^u,$or ${\lambda }_{4,\text{O2}}^{u+},$and these upstream signals will be received by receivers Rx1(M2), Rx2(M3), Rx2(M2), Rx3(M4), Rx3(M2), Rx4(M1), and Rx4(M2), respectively, as shown in Fig. 5(b). So the upstream capacity for ODN2 can reach 70 Gb/s as well. Meanwhile, ONUs in ODN3 with transmitter wavelength ${\lambda }_{4,\text{O3}}^u$must be tuned to ${\lambda }_{3,\text{O3}}^{u+}$, and ONUs in ODN4 with transmitter wavelength ${\lambda }_{3,\text{O4}}^u$must be tuned to ${\lambda }_{4,\text{O4}}^{u+}$, in order to avoid collision of upstream wavelengths in the same receiver. These two wavelengths, ${\lambda }_{3,\text{O3}}^{u+}$and${\lambda }_{4,\text{O4}}^{u+}$, are received by receivers Rx3(M3) and Rx4(M4), respectively. With such a wavelength arrangement in upstream, there is only one wavelength going to each receiver at OLT, except Rx1 in module M2 where both ${\lambda }_{1,\text{O2}}^u$and${\lambda }_{1,\text{O2}}^{u+}$ come from ONUs in ODN2. As each WDM demultiplexer in the transceiver module has 200 GHz channel spacing, so tunable receiver is not needed for load balancing.

 

Fig. 5 Load balancing in flexible TWDM PON with fixed receiver at OLT. (a) shows the downstream wavelength assignment and (b) shows the upstream wavelength assignment. With such wavelength assignment, 70Gb/s aggregated bandwidth can be provided to ODN2 in both downstream and upstream.

Download Full Size | PDF

In addition to load balancing, the same mechanism can be utilized to provide resilience against OLT transceiver failures. For example, when OLT transceiver module M2 fails, the wavelengths of OLT transceiver module M1 can be tuned from the original wavelength set, ${\lambda }_1^d/{\lambda }_1^u,{\lambda }_2^d/{\lambda }_2^u,{\lambda }_3^d/{\lambda }_3^u$and${\lambda }_4^d/{\lambda }_4^{u+},$ to a shifted wavelength set ${\lambda }_1^{d+}/{\lambda }_1^{u+},{\lambda }_2^{d+}/{\lambda }_2^{u+},{\lambda }_3^{d+}/{\lambda }_3^{u+}$and${\lambda }_4^{d+}/{\lambda }_4^{u+}$, so that transceiver M1 can serve both ODN1 and ODN2. Therefore, OLT transceiver protection can be achieved in the flexible TWDM PONs.

2.3 Power saving at OLT

In addition to pay-as-you-grow and load balancing, the proposed architecture also allows significant power saving at OLT. As the traffic load in a single ODN reduces, a conventional TWDM PON will reduce the number of activated wavelengths, and shut off some of the transceivers in OLT. However, as long as there are any active users in an ODN, at least one wavelength pair must remain active for this ODN. With our proposed architecture, if the traffic load is very small, we could shut off all the wavelengths in OLT except one wavelength, e.g. ${\lambda }_1^{d+}$(M1), as shown in Fig. 6. This wavelength is shifted by 100 GHz from the original wavelength ${\lambda }_1^d$(M1), so that this downstream wavelength will go through the 4 × 4 power splitter and reaches all the users in 4 ODNs. In upstream, all the ONU transmitters will be tuned to wavelength ${\lambda }_1^{u+}$, so that the upstream wavelength from any ODNs goes through the 4 × 4 power splitter and reaches the receiver Rx1 in module M1. In comparison, for conventional TWDM PON architecture operating in power saving mode, 4 transceivers must be kept active to serve 4 different ODNs. Hence, the flexible TWDM PON can achieve more power saving at OLT.

 

Fig. 6 Power saving in flexible TWDM PON. If the traffic load in a flexible TWDM PON is low, a single OLT transceiver (TRx1 in module M1) can serve multiple ODNs, achieving significant power saving. In comparison, at least one transceiver in each OLT module must remain active in a conventional TWDM PON operating in power saving mode.

Download Full Size | PDF

2.4 Experimental test bed

To demonstrate the feasibility of the proposed flexible TWDM PON, an experimental test bed is built with the architecture shown in Fig. 2. For downstream, four 10 Gb/s EMLs (electroabsorption modulated laser) with wavelengths on ITU grid at 1587.88, 1589.57, 1591.26 and 1592.95 nm are used at OLT side. These EMLs can be thermally tuned by 100 GHz to wavelengths at 1588.73, 1590.41, 1592.10 and 1593.79 nm, respectively. At ONU side, a thermally tuned optical filter is packaged inside ROSA to select one of the downstream wavelengths, and a three-section DBR laser with external modulation is used as the tunable transmitter for 10 Gb/s upstream transmission. The DBR laser wavelength can be tuned from 1530 to 1540 nm. With these upstream and downstream wavelengths, our TWDM PON can coexist with all the legacy PON systems. Figure 7 shows the test results for thermal tuning of EMLs at OLT and current tuning of DBR lasers at ONUs. The tuning speeds for EMLs and DBR lasers are about 80 ms and 50 ns respectively (results include the respond time of the driving circuit inside the transceivers). Since each wavelength pair (for downstream and upstream) in the TWDM PON serves only a few users in a single ODN, the traffic could vary significantly within a short period of time due to the nature of the self similar traffic from end users [16]. The 50 ns tuning speed of DBR laser is able to support fast load balancing within the same ODN for packet switching on a time scale of 125 μs (i.e. GPON/XG-PON frame cycle). On the other hand, the aggregated traffic load from an ODN is the sum of the traffic generated by a large number of end users (e.g. 64 users), so the variation of the aggregated traffic in a single ODN is slower and the load balancing among different ODNs can be supported by the slower lambda switching. Slow tuning speed of thermal tuned EML, on 100 ms time scale, provides the needed lambda flow among different ODNs in the flexible TWDM PON.

 

Fig. 7 Tuning speed of EMLs and DBR lasers. Two curves in the each diagram represent the received powers in two adjacent channels (100 GHz spacing) when the wavelength of the EML for downstream (DBR for upstream) is switched from λ^d to λ^d+ in downstream (λ^u to λ^u+ in upstream).

Download Full Size | PDF

The transmission performance for all the 10 Gb/s downstream and upstream channels are tested in the flexible TWDM PON, and the measurement results for bit error rates (BER) are shown in Fig. 8. After 20km single mode fiber, the receiver sensitivity is better than −28 dBm (at BER = 10⁻³) for downstream and −36 dBm for upstream (at BER = 10⁻⁴). Compared to the back-to-back case, the power penalty is less than 1 dB for both upstream and downstream transmission. With a booster amplifier at OLT, the transmitted power in each downstream channel is 10 dBm. For upstream, the transmitted power from each ONU is about 3 dBm. Hence, the power budget for downstream and upstream is more than 38 dB, enough to support 20km reach and 1:64 split after the hybrid AWG/splitter.

 

Fig. 8 Bit error rates of 10 Gb/s downstream (top) and upstream (bottom) transmissions for back-to-back and after 20km standard single mode fiber in the flexible TWDM PON test bed.

Download Full Size | PDF

3. Pluggable optical transceiver modules

For large scale deployment of TWDM PONs in the near future, technology development and engineering solutions for optical transceiver modules must be made commercially viable for manufacturability and cost effectiveness. At OLT side, performance and footprint are very important, so the integration of both electronics and optics is a primary task. On the other hand, transceiver module design at ONU side is simpler but low cost is the utmost goal. Meanwhile, good performance is required for ONU transceivers under cost constraint. Hence, trade-off between cost and performance is necessary for ONU transceivers. In addition, built-in RSSI (Receiver Signal Strength Indicator) for both OLT and ONU transceivers is necessary for automatic wavelength alignment in TWDM PONs. With these requirements in mind, we developed OLT transceiver module in enhanced CFP package and ONU transceiver module in SFP+ package for cost effect deployment of TWDM PONs.

3.1 OLT and ONU transceiver module development

Figure 9(a) illustrates the design of the 4-channel OLT transceiver module. For the transmitter, there are 4 EMLs, each modulated by a 10 Gb/s data stream. The outputs of EMLs at different wavelengths are then multiplexed by a low-loss multiplexer and amplified by an L-band EDFA. At the receiver side, optical signals from ONUs in 4 different wavelength channels are first amplified by a C-band EDFA and then separated by a demux before being detected by burst-mode APD ROSAs. Limiting amplifiers (LA) following the ROSA further boost the received signals for data recovery. To achieve bidirectional transmission on a single fiber, a WDM filter combines/separates the upstream and downstream wavelengths. A microcontroller (MCU) is also included in the transceiver module for control and monitoring purpose. It sets the operation conditions of various optical and electronic components in the module, and performs monitoring functions such as transmitter output power monitoring, receiver signal power measurement, EDFA gain control and loss-of-signal alarm processing. Figure 9(c) shows a picture of the pluggable transceiver module in enhanced CFP package.

 

Fig. 9 Pluggable optical transceiver modules for TWDM PONs.

Download Full Size | PDF

Figure 9(b) illustrates the design of the tunable ONU transceiver module. For transmitter, it uses a thermally tuned DFB laser directly modulated by a burst-mode laser diode driver at 2.5 Gb/s. The DFB laser can emit relatively large power but can only be tuned over a few channels with 100 GHz spacing and its tuning speed is about 80 ms. At the receiver side, a tunable optical filter (TF) is packaged inside the ROSA together with an APD photodiode and a transimpedance amplifier (TIA). The tunable filter can be thermally tuned to one of the 4 downstream wavelengths in L-band, and its tuning speed is on the order of 100 ms. The tunable filter provides over 25 dB channel isolation to ensure good performance. Automatic gain control for APD, TIA and LA results in a large dynamic range for the received optical signal. Figure 9(c) shows a picture of the tunable ONU transceiver module in SFP+ package.

3.2 OLT and ONU transceiver testing results

Figure 10(a) shows the eye diagrams of the OLT transmitters with 10 dBm output power and 9 dB extinction ratio. The optical spectrum of the downstream signals is plotted in Fig. 10(b), showing four channels on ITU grid (1608.33, 1609.19, 1610.06 and 1610.92 nm). The OLT receiver sensitivity in burst mode at 2.5 Gb/s is better than −36dBm at BER = 10⁻³. Figure 10(a) also shows the eye diagrams of the ONU transmitter with its wavelength tuned to 4 upstream wavelength channels; its average transmitted power is 4 dBm and extinction ratio is better than 9 dB. The optical spectra of the upstream signals in different channels (with wavelengths at 1535.82, 1536.71, 1537.41, 1538.19 nm) are plotted in Fig. 10(c). ONU receiver sensitivity at 10 Gb/s is better than −26dBm at BER = 10⁻⁴, and the receiver overload is more than 0 dBm using automatic gain control.

 

Fig. 10 Testing results of OLT and ONU transceiver module. (a) Eye diagrams from 4 OLT transmitters (top) and an ONU transmitter (bottom) tuned to 4 different wavelengths with 100 GHz channel spacing. (b) Optical spectrum of the transmitted signals from OLT transceiver module; (c) Optical spectra of an ONU transmitter tuned to 4 different upstream channels.

Download Full Size | PDF

4. System demonstration with pluggable transceiver modules

For performance verification and network throughput testing, OLT and ONU transceiver modules are plugged in the flexible TWDM PON system test bed. The bit error rates of the downstream and upstream transmissions are shown in Fig. 11 for all the channels at back-to-back, after 20km and 40km single mode fiber. With these optical transceivers, 36 dB power budget is achieved for both downstream and upstream transmissions.

 

Fig. 11 Bit error rates for 10 Gb/s downstream (top) and 2.5 Gb/s upstream (bottom) transmissions at back-to-back, after 20km and 40km standard single mode fiber.

Download Full Size | PDF

In TWDM PON system, ONU must automatically align its transceiver wavelengths (for both the transmitter and the receiver) to the right channel. This can be done through an embedded control channel between OLT and ONU. In our system, when an ONU is powered on, it will scan the downstream wavelengths by tuning its receiver. The tunable receiver can lock to a specific downstream channel by monitoring the received power through the RSSI built inside the ONU transceiver. For upstream wavelength alignment, the tunable ONU scans its transmitter wavelength, while OLT monitors the received power through RSSI in the OLT transceiver module. Figure 12 shows the RSSI values obtained at OLT when ONU scans its wavelength (controlled by DAC value with roughly 4 pm per step) across upstream channel 3. By finding the maximum value of RSSI, OLT can control the ONU transmitter wavelength with wavelength setting command sent through the control channel embedded in the downstream PLOAM (Physical Layer Operations, Administration and Maintenance) message. Initially during ONU activation, ONU transmitter wavelength might not be set in the middle of the channel due to the limited accuracy (about ± 1dB) of RSSI from OLT transceiver, but with such RSSI accuracy, ONU wavelength can be tuned within the 1 dB passband of the channel. This coarse wavelength alignment ensures successful upstream connection and ONU activation process. After ONU enters normal operation phase, the RSSI value can be recorded for each burst from this ONU. By averaging a large number of RSSI values, RSSI accuracy can be improved significantly. Eventually, by monitoring the RSSI value for a large number of upstream bursts, the ONU wavelength can be fine tuned to the center of the upstream channel.

 

Fig. 12 Measured RSSI value from OLT transceiver when ONU wavelength is tuned across upstream channel 3. The ONU wavelength is set by a DAC (digital to analog converter) with each DAC step corresponding to a change of 4 pm in ONU wavelength.

Download Full Size | PDF

After ONU wavelength alignment and activation, network throughput is evaluated in our system test bed. For upstream, 5 ONUs transmit upstream signals in the same wavelength channel with time division multiple access, and each ONU is loaded with Ethernet packets of random length (64-1518 bytes). 2.3 Gb/s aggregated throughput in a single upstream channel is achieved without any packet drop. Meanwhile, for downstream, Ethernet packets with random length (64-1518 bytes) are transmitted by the OLT at 10 Gb/s in each downstream channel. No packet drop is observed in the downstream with traffic load up to 1.0 Gb/s for each ONU (limited by the gigabit Ethernet port in the ONUs).

5. Conclusions

In this paper, a flexible TWDM PON architecture is presented that allows pay-as-you-grow deployment of OLT transceivers for smooth bandwidth upgrade, achieves load balancing among different ODNs, provides channel protection for OLT transceivers, and supports selective OLT sleep for significant energy saving. The flexible TWDM PON requires only narrowly tuned transmitters at OLT with 100 GHz tuning range, and passive components (hybrid AWG and splitter) in ODN. For cost effective deployment of flexible TWDM PONs, integrated OLT transceiver in enhanced CFP module and low-cost tunable ONU transceiver in SFP+ module are developed for the first time. System transmission experiments demonstrate more than 36 dB power budget under FEC limit for error free performance, and network performance testing shows 2.3 Gb/s aggregated throughput in upstream and 1.0 Gb/s throughput per ONU in downstream.