Open Access
Add to BibSonomy
Abstract
Using a dual-wavelength source, a single optical signal is distributed over two wavelengths. This approach is used to reduce power excursions due to optical circuit switching in ROADM systems. In a multi-hop optical transmission system with 100Gbps PM-QPSK signals switched over five ROADMs and 265 km of single mode fiber (SMF), power excursions are kept within ± 0.2 dB using dual-wavelength sources. A cumulative distribution function (CDF) of channel power excursions is generated from measurements of over 100 random channel loadings for four different channel loading plans. Using dual-wavelength sources, power excursions of up to ± 6 dB are reduced below ± 1.5 dB for 99% and ± 0.5 dB for 71.4% of the cases while switching up to 30 channels.
© 2017 Optical Society of America
1. Introduction
The continuous growth of Internet applications, including for example cloud computing and video rendering and network scale resources as a service, is rapidly increasing the data traffic in optical transmission systems, and as a result is motivating more efficient and application aware capabilities. Dynamic wavelength-division-multiplexed (WDM) systems in which resources are allocated with real-time wavelength switching is a long sought-after such capability [1–4]. Traditional commercial WDM systems remain quasi-static, with wavelengths being provisioned once to meet the peak traffic requirements and remaining unchanged [5]. To support dynamic wavelength route adaptation, reconfigurable optical add-drop multiplexers (ROADMs) in WDM systems need to be rapidly controlled in response to changing channel configurations.
A key challenge in converting today’s quasi-static WDM systems into real-time wavelength routed WDM systems is the power dynamics that arise during optical circuit switching operations. The optical power dynamics due to uncontrolled or constant power erbium-doped fiber amplifiers (EDFAs) can be dramatically reduced by using automatic gain controlled (AGC) EDFAs, which are designed to maintain constant average gain for all active channels [6]. However, optical power dynamics still occur in AGC EDFAs during wavelength reconfiguration and can grow in cascades. There are two types of optical power dynamics in the EDFA—transient effects and optical power excursions. EDFA transient effects are fast power over and undershoots that arise from sudden changes in input power due to wavelength reconfiguration or upstream fiber cuts, since all channels present are amplified simultaneously and share the available gain. Notice transient effects occur even without wavelength gain dependency. Transient effects have been largely solved by fast feed-forward control which effectively and rapidly mitigates fast power over and undershoots [7, 8]. Nevertheless, after the transient effect, long lived power excursions that result from the wavelength dependent gain of EDFAs persist after wavelength reconfiguration operations. Power excursions are persistent power differences on WDM channels after versus before wavelength reconfiguration. These power dynamics can grow in magnitude over cascaded EDFAs and lead to an optical signal to noise ratio (OSNR) degradation at low power or high optical fiber nonlinearity impairments at high power. Though the cumulative effects of these power dynamics can be relieved by per-channel power adjustment at wavelength selective switches (WSSs) or variable optical attenuators (VOAs) in the ROADM nodes, this requires minutes of tuning per wavelength, as in recently reported in commercial field trials [9]. Control algorithms for such power adjustments are even more complicated in a mesh or ring topology, because the power excursions can propagate through the network passed from channel to channel through the optical power dynamics. Multiple parallel power tuning adjustments in a network increase these effects and can lead to instability. Using sequential adjustments, on the other hand, leads to long provisioning delays [10].
Machine learning (ML) methods have been investigated to mitigate power excursions, looking at power divergence and OSNR estimates. A ML model was developed to recommend a channel with minimized power divergence among 24 given channels in a system with three cascading EDFAs [11, 12]. Previous research also implemented a case-based-reasoning ML approach to maintain the OSNR, but required a high number of measurements to improve the OSNR by 1 dB [13].
Analytical models have also been developed to predict and minimize power excursions. The wavelength dependent gain was estimated by measuring the fully loaded WDM channel gain and single channel gain, and generated an estimated gain spectrum to perform wavelength assignment in order to minimize power excursions [14, 15]. 5%-15% power excursion reductions were reported in a single-hop end-to-end network with five cascading EDFAs.
Using a fast tunable laser to distribute a single signal over two or more wavelengths with different wavelength dependent gains is also a promising way to minimize the power excursions. By adjusting the duty ratio of the fast tuning between two wavelengths, less than 0.1 dB power excursion was achieved on measured channels in a three cascading EDFA single-hop system [16]. However, its data transmission performance in a multi-hop ROADM transmission system with dynamic wavelength reconfiguration has yet to be explored. Recent work investigated dual-wavelength sources in a transmission system over 8 amplified fiber spans with the 255 km transmission fiber [17]. A dual-wavelength source was used to modulate a 100 Gb/s PM-QPSK signal and successfully mitigated power excursions to be under 0.2 dB. However, only a single dual-wavelength source was used in this work. The effectiveness of using multiple dual-wavelength sources in a multi-hop ROADM system under optical circuit switching operations has not been studied. Note that different wavelengths have different propagation speed in the transmission fiber, and therefore a propagation delay may occur between different wavelengths at the receiver and the accumulated impairments will differ. A guard band is needed at the receiver, considering the propagation delay as well as the tuning transition. We also note that one can use a dual-wavelength source in which two continuous-wave (CW) signals are modulated by the same modulator at the source and only one is detected at the receiver while the other is thrown away. It eliminates the need for guard bands but increases the total average signal power by 3 dB. For a CW dual-wavelength source, the duty ratio adjustment becomes a real change in optical power, compromising the performance at one of the wavelengths. Nevertheless, results obtained using the fast tuned dual-wavelength source are generalizable to the CW dual-wavelength source case.
From a transmission perspective, nanosecond speed switching has been studied for several different modulation formats, including coherent detection systems. A digital supermode distributed Bragg reflector (DS-DBR) laser was used as a local oscillator in a 112 Gb/s PM-QPSK back-to-back system containing 24 channels. The tunable laser was able to switch to any one of the 24 channels in less than 130 nanoseconds (ns), and demodulate 5-microsecond (µs) optical bursts with a 1.7 to 3.2 dB OSNR penalty [18]. A tunable sampled grating distributed Bragg reflector (SG-DBR) laser was used to switch between a pair of wavelengths, modulated with different modulation formats. The guard band that resulted from the tuning transition was evaluated to be 4 ns, 7 ns, and 40 ns, for 10.7 Gb/s NRZ-OOK, 10.7 Gb/s DPSK and 21.4 Gb/s DQPSK respectively [19].
Using dual-wavelength sources in ROADM systems for wavelength reconfiguration and optical circuit switching is a promising direction to minimize power excursions and reduce or eliminate time-consuming power adjustments that compensate for these excursions. However three questions remain unsolved under optical circuit switching operation: (i) how much improvement that dual-wavelength sources can achieve, regarding the reduction of power excursions and the amount of time for power adjustments; (ii) whether the performance of dual-wavelength sources differs under different channel loadings (e.g. number of active channels, active channel locations, and channel hop count); (iii) what is a strategy to use dual-wavelength sources together with single-wavelength sources for better spectral utilization.
To evaluate the effectiveness of dual-wavelength sources under optical circuit switching operation, we study power excursion mitigation with a bisection wavelength assignment algorithm, and evaluate the performance of two different tuning approaches-blind tuning and optimal tuning. Results are compared with single-wavelength source optical circuit switching. First, the transmission performance of dual-wavelength sources is evaluated with 100Gbps PM-QPSK signals. Second, the performance of dual-wavelength source based optical circuit switching is evaluated under different channel configurations, and the cumulative distribution function (CDF) of channel power excursions with different channel loading plans are compared. Finally, we study the power excursion mitigation and the SE trade-off, and propose a strategy to transition from dual-wavelength sources to single wavelength sources as the channel loading increases, in order to preserve spectral efficiency. The organization of the paper is as follows. In section 2, a mathematical model of power excursions is provided, and the principle of the dual-wavelength source approach is illustrated. The experimental setup of a 90-channel 5-hop ROADM system is presented in section 3. Experimental results are discussed in section 4, and we conclude our findings in section 5.
2. Principal of channel power excursions and dual-wavelength sources
EDFAs are deployed in ROADM networks between ROADMs to compensate for fiber attenuation and component losses. AGC EDFAs are commonly used in ROADM systems and keep the ratio of the total output power to the total input power constant (i.e. constant average gain). The constant average gain is achieved by collecting the continuous readings of the total optical power from output and input photodetectors and accordingly tuning the amplifier pump power. The control speed depends on the control loop and the amplifier response, and is usually on the order of tens of microseconds [6]. An EDFA has wavelength dependent gain on the order of ± 0.5-1 dB, which is minimized at the design point of the gain flattening filter (GFF) and can vary over the operating range of the EDFA. Effects such as spectral hole burning (SHB) in the short wavelength range of the gain spectrum further contribute to the wavelength dependence. Furthermore, the wavelength dependent gain is often deliberately modified in order to compensate for effects such as stimulated Raman scattering (SRS) in the transmission fiber. When wavelength reconfiguration modifies the location and power of wavelengths at the EDFA input, the EDFA responds to the wavelength dependent power changes to meet the AGC condition, giving rise to power excursions [20]. Neglecting gain dependent tilt and other secondary gain dependent effects, all channels in an amplifier experience the same power excursion, which can be approximated by the following analytical equations:
GT is the target gain setting for the AGC EDFA, which is equal to the mean gain of all channels and is set to match the span loss or set for a desired channel output power. Gj is the wavelength dependent gain for channel j. Pi,j and Po,j denote the input power and output power for channel j. N and N' denote the number of channels input to the EDFA before and after wavelength reconfiguration. To quantify the power excursion on a certain channel k during wavelength reconfiguration, we use Pi,k and Po,k to denote input power and output power for channel k before wavelength configuration [Eq. (1)], and use P'o,k to denote output power for channel k after wavelength reconfiguration [Eq. (2)]. Before wavelength reconfiguration, channels are at a stable state (i.e. AGC condition is maintained) such that their mean gain equals to the EDFA target gain [Eq. (1)]. When wavelength reconfiguration occurs with channels appearing on different wavelengths or changing channel input power, the AGC condition no longer holds because input power Pi,j changes to P'i,j and the number of channels changes from N to N' [Eq. (2)]. To maintain the AGC condition, all channel gains need to be tuned by ΔG, resulting in power excursions ΔP = ΔG (in decibels) as shown in Eq. (3). There are three basic rules to quantify the amplitude of power excursions: (i) power excursions are determined by the product of the wavelength dependent gain and the input power; (ii) the sign of power excursions depends on the average gain of active channels and the average gain of added channels. If the average gain of the added channels is larger, the AGC EDFA decreases the pump power to maintain constant average gain, leading to negative power excursions. Vice versa, a smaller average gain for the added channels leads to positive power excursions; (iii) the amplitude of the power excursions is impacted by the number of active channels before and after wavelength reconfiguration. For example, with a given the number of new channels, the power excursions will be reduced with increasing number of unchanged active channels. Note that in a multi-hop ROADM system containing multiple EDFAs, the power excursions should also consider the cascading effect, potentially leading to large OSNR degradations and worse BERs.
Figure 1(a) depicts the power excursions that occur due to adding a new channel using a single wavelength source, with the AGC EDFA being operated at 17.9 dB gain. The black curve represents the EDFA’s wavelength dependent gain with only the wavelength at λ1 active and the gain at λ1 (solid blue arrow) is 17.9 dB. Then wavelength reconfiguration happens such that a new channel with wavelength λ2 (solid red arrow) is added into the path of the EDFA. Before the AGC operation responds, both channel gains are unchanged; however, the mean gain is now at 17.56 dB, owing to the lower gain of λ2 (17.20 dB). Within tens of microseconds, the AGC operation adjusts the pump power such that both channel gains are increased by 0.34 dB (shown as dashed arrows) to a new stable state with a new wavelength dependent gain profile (green curve), resulting in 0.34 dB power excursions on both channels. Note that at this new stable state, the AGC condition is held such that the average gain of the two channels still equals to 17.9 dB. Also note that this effect grows in cascade with the worst case occurring in networks with power leveling on each hop in which case the excursion (in decibels) is multiplied by the number of hops [21].
Fig. 1 EDFA AGC introduces power excursions: (a) positive channel power excursions occur when channel at λ2 is added, due to the gain changes to maintain AGC condition; (b) by adding channels at both λ2 and λ3, power excursions cancel out, and gain profile is unchanged.
Instead of adding a new channel with a single wavelength, Fig. 1(b) shows that power excursions are minimized by adding a new dual-wavelength signal. The same channel at wavelength λ1 is active (blue arrow), and two channels with wavelength λ2 (red arrow) and λ3 (magenta arrow) are added at the same time. Since the gain of λ2 is less than that of λ1 and the gain of λ3 is larger than that of λ1, with the appropriate choice of λ2 and λ3, ΔG = 0 dB is achieved and no power excursion occurs. Since it is the product of the wavelength dependent gain and the input power that determines the power excursion, if an exact pair is not available to balance out the excursion, then the relative powers can be adjusted to compensate for the power excursion. However, changing the channel powers can impact their transmission performance. Also using two signals doubles the number of lasers, when only one is needed. A solution to both issues is using a dual-wavelength signal, which is realized by nanosecond speed wavelength tuning between two wavelengths. As long as the tuning period is faster than the amplifier response time, the amplifier will perceive the dual-wavelength signal as two separate and constant signals. A tuning period of approximately 10 µs is fast enough since the response of the AGC EDFA is on the order of tens of microseconds. In this case, rather than adjusting the channel powers to balance the power excursions, the duty ratio can be varied so that the average power seen by the amplifier is offset by the required amount. A dual-wavelength source can be implemented either using a transceiver with a fast tunable laser source or a transceiver with two slow tunable laser sources coupled using a fast 2 × 1 optical switch (see Section 3). By appropriately selecting a wavelength pair and tuning their duty ratios, channel power excursions can be minimized, and OSNR and BER levels across all channels can be maintained.
3. Experimental setup
Figure 2 shows the experiment setup. The 5-ROADM network diagram is shown in Fig. 2(a), which includes 4 fiber spans with various lengths totaling 265km. Two dual-stage AGC EDFAs are set at 18 dB gain in each span to compensate for the loss of the transmission fiber and the ROADM, and variable attenuators are used to increase the loss of transmission fiber and the loss of the ROADM node to 18 dB. Each ROADM contains 1 × 2 and 1 × 4 WSSs with per-channel VOA control and optical channel monitoring (OCM) modules. The first ROADM only has an ‘add’ port for adding channels and the last ROADM only has a ‘drop’ port. The other three ROADMs have both add and drop functionality. Each ROADM drop port is connected to an OCM for power measurements. A 100 Gb/s PM-QPSK transmitter with a 90-channel 50-GHz spaced DWDM source is connected to a 1 × 4 coupler, allowing channels to be added into different ROADMs.
Fig. 2 Experimental setup. (a) 5-ROADM network separated by 4 spans with 2 AGC EDFAs on each span; (b) 100G PM-QPSK transmitter diagram (TX); (c) 100G PM-QPSK receiver diagram (RX).
The 100 Gb/s PM-QPSK transmitter setup is shown in Fig. 2(b). The source includes a 90-channel WDM source to bring up active channels, an SG-DBR laser with nanosecond tuning speed for fast laser switching, and a pair of integrable tunable laser assembly (ITLA) lasers coupled by a 2 × 1 optical space switch with nanosecond switching speed. Dual-wavelength sources can be implemented by either the fast tunable SG-DBR laser or two slow tunable ITLA lasers coupled using the 2 × 1 switch. The composite source modulates 50 Gb/s QPSK signals (i.e. 25 GHz baud rate, with 53 symbol decorrelation between I and Q) and 100 Gb/s PM-QPSK signals are generated by splitting and recombining signals with 258 symbol delays. The combined signal is then sent into the ROADM system to investigate the signal performance penalty due to switching and power excursions. The coherent receiver setup is shown in Fig. 2(c). A broadband ASE noise generator is introduced before the PM-QPSK receiver to measure the BER for different OSNR values. A tunable bandpass filter (BPF) selects a channel for BER analysis. The receiver also consists of an ITLA local oscillator (LO), four 28 GHz photodetectors, and a 4-channel digital oscilloscope (DPO) with 100 Gsamples/s. Offline digital signal processing (DSP) is performed on 1 million symbols of 20 µs duration for each channel (with a BER precision at 5 × 10−7.) Note that for analyzing the BER of a dual-wavelength signal, the LO is tuned to one of the two wavelengths at a time, although using a fast tunable LO to select between the two wavelengths has been shown previously [18].
The initial experiment setting is as follows. To investigate the BER performance under power excursions using dual-wavelength switching and single-wavelength switching approaches, the OSNR is set at 15 dB with a BER around 2 × 10−3. All AGC EDFA target gains are set to 18 dB, but each EDFA has different wavelength dependent gain. Figure 2 inset shows the wavelength dependent gain of two EDFAs of the first span. The Fig. 2 inset shows the wavelength dependent gain of two EDFAs of the first span. The first EDFA is set with a small negative tilt (−0.3 dB as shown in the inset) to compensate for the stimulated Raman scattering (SRS) in the transmission fiber. The second EDFA is mainly used to generate wavelength gain dependencies to study power excursions. EDFAs with similar architectures and gain flattening filters (e.g. same make and model) will have similar wavelength dependent gain. With 90-channel WDM input, the VOA attenuation values are set to ensure 0 dBm launch power (i.e. input power into the transmission fiber) per channel, with a 19.5 dBm output power at each EDFA. These attenuation values are stored as a reference. This setting only compensates for the channel power divergence due to wavelength dependent gain and SRS. The power excursions still occur during wavelength reconfiguration. In the following experiments, we choose power excursion margins of ± 1.5 dB for general channels and ± 0.5 dB for quality-sensitive channels, corresponding to the pre-BER level of 1 × 10−2 and 3 × 10−3. For different channel configurations, the launch power of active channels is tuned to be 0 dBm per channel, and new channels are added using the pre-stored VOA values.
4. Experimental results and discussions
4.1 100 Gb/s PM-QPSK transmission performance using dual-wavelength sources
We evaluated the effectiveness of the dual-wavelength source by both using a nanosecond switching speed tunable SG-DBR laser (fast tunable laser approach) and two slow tunable ITLAs coupled by a 2 × 1 nanosecond switching speed optical switch (slow tunable laser approach). Both approaches have similar performance and only the results of the slow tunable laser approach are shown in the experimental results. Figure 3(a) shows the measured back to back BER and the BER after 4-span transmission (265km) as a function of OSNR at the receiver. Figure 3(b) shows the cascading effect over 4 ROADM spans of an active channel with the wavelength of 1547.1nm, when a new channel with the wavelength of 1563.1nm is added into the network. Power excursions at the receiver when using a single-wavelength source and dual-wavelength source are shown in Fig. 3(c). For the single-wavelength source, the new channel wavelength is fixed at 1563.1nm, while the active channel wavelength moves from 1535.1nm to 1559.1nm and its power excursion and BER are correspondingly measured. For the dual-wavelength source, the dual-wavelength signal is tuned between 1530.7nm and 1563.1nm, while the active channel moves from 1535.1nm to 1559.1nm. To compensate for the tuning transition and the propagation delay of the dual-wavelength signal, a guard band of 276 ns is set for each tuning event at the receiver. The guard band introduces the overhead in the transmission system. However, the overhead can be significantly reduced by decreasing the switching frequency. With a tuning period of 10µs (which is still much faster than AGC response time), the overhead is less than 3% for 100Gbps PM-QPSK signals. The BER measurements are performed for all data outside of the guard band. Power excursions are measured with an initial 50:50 duty ratio (green diamond) and an optimal duty ratio (blue star). The single-wavelength source yields more than 0.5 dB power excursions at all of the measured channels. Using the dual-wavelength source with optimal duty ratios, power excursions are reduced to less than 0.2 dB for all measured channels. The dual-wavelength source with an initial 50:50 duty ratio also outperforms the single-wavelength source, ensuring power excursions less than 0.5 dB for all measured channels. Figure 3(d) illustrates the BER performance of an active channel (already provisioned) at different wavelengths, when a new signal is introduced into the system using a dual-wavelength source (blue star in Fig. 3(d)) or a single-wavelength source (red cross in Fig. 3(d)). Introducing a new signal using the dual-wavelength source effectively maintains the active channel’s BER below 3 × 10−3 and keeps its OSNR above 14 dB. This BER is below the FEC limit 3.8 × 10−3, and the data rate accounting for FEC overhead of 12% is 112 Gb/s [22].
Fig. 3 Transmission performance using a dual-wavelength source. (a) BER vs. OSNR of back to back and after 265 km (4 span) transmission; (b) cascading power excursions on active channel 45, when a new single wavelength channel 85 is added; (c) power excursions vs. active channel location using a dual-wavelength source and a single-wavelength source; (d) BER and OSNR vs. active channel.
4.2 Dual-wavelength source based optical circuit switching and wavelength reconfiguration
The preceding transmission experiment shows that using a dual-wavelength source, the power excursion on the active channel is minimized and the BER and OSNR performance is maintained, with only two active channels as an important corner case. In a multi-hop ROADM system under optical circuit switching operation, frequent wavelength reconfiguration can occur within a short period of time. Moreover, different channels have different physical routes, making the power excursion mitigation more difficult. For example, there is one active channel at wavelength λ1 with the route from ROADM1 to ROADM2, and another active channel at wavelength λ2 with the route from ROADM4 to ROADM5. If λ1 and λ2 have different gain spectra, and wavelength reconfiguration requires a new channel to be added from ROADM1 to ROADM5, then the excursions due to the presence of these two wavelengths cannot be simultaneously canceled. To avoid excursions on the first active channel at λ1, the aforementioned dual-wavelength source can be used, but the new dual-wavelength signal would inevitably generate power excursions on the second active channel with a wavelength λ2. Therefore, under optical circuit switching, the power excursions depend on not only the active channel locations but also the physical routes of both new channels and active channels. To evaluate the effectiveness of dual-wavelength optical circuit switching, instead of quantifying the power excursion of a certain channel, we explore its effectiveness to minimize the maximum power excursion among all active channels. In the following experiments, we first evaluate both dual-wavelength sources and single-wavelength sources with four specific channel loadings. Secondly, dual-wavelength source based optical circuit switching is evaluated using the cumulative distribution function (CDF) of the measured power excursions with random channel loadings resulting from four different channel loading plans. Notice that each dual-wavelength signal is generated using two stable lasers (from the 90-channel comb source in Fig. 2(b)) for which the relative powers are adjusted corresponding to the dual laser switched source. The experimental results will not be affected since these signals lead to the same power excursion effect (which occurs over long time scales) as the fast tunable laser (or a pair of ITLA coupled with a 1 × 2 switch) based dual-wavelength sources.
Figure 4 shows the power excursions generated using dual-wavelength sources and single-wavelength sources. Four channel loadings are evaluated, which are (a) short-hop, (b) long-hop, (c) uniform distribution, and (d) distribution on long wavelengths, shown in Fig. 4(a)-4(d) respectively. Seven initial channels with random add and drop ports are loaded to model the case of an initially lightly loaded system. The y-axis represents the absolute values of maximum power excursions measured on all active channels and the x-axis represents the number of new channels added at a time.
Fig. 4 Comparisons of dual-wavelength sources and single-wavelength sources. The x-axis represents the number of new channels added at a time. (a) Short-hop traffic dominated; (b) long-hop traffic dominated; (c) active channels are uniformly distributed; (d) active channels are distributed on long wavelengths.
Notice that in the dual-wavelength case, a dual-wavelength channel comprises of two wavelengths. For single-wavelength sources, four wavelength assignment algorithms are evaluated: (i) leftmost: choose the shortest available wavelength; (ii) rightmost: choose the longest available wavelength; (iii) centermost: choose the most centered available wavelength; (iv) random. For dual-wavelength source wavelength assignment, a bisection wavelength assignment algorithm is implemented as follows: 90 wavelengths are split into two groups (i.e. wavelength 1-45, and wavelength 90-46) and the wavelengths of dual-wavelength channels are selected from the sides to the center (i.e., if available, channel 1 and 90 forms the first channel, and wavelength 45 and 46 forms the last channel). The bisection wavelength assignment algorithm works well for these EDFAs as there is not much structure to the ripple. The effectiveness of the bisection wavelength assignment depends on the ripple function of each EDFA. If all the EDFAs have similar gain characteristics (i.e., gain ripple functions), the structure of the cascaded gain ripple function is simpler since it is the addition of multiple identical gain characteristics. For an EDFA gain spectra containing more frequency components, an N-section wavelength assignment algorithm can be used.
The power excursion mitigation by using both blind tuning and the optimal tuning is evaluated. In blind tuning, new dual-wavelength channels are always added with 50:50 duty ratio; while for optimal tuning, the duty ratio of new dual-wavelength channels are adjusted to ensure that the maximum absolute power excursion of all active channels is minimized. Additional tuning time is needed for optimal tuning to converge on the optimal duty cycle. The amount of tuning time per cycle depends on the speed of the feedback messages from the OCM and the WSS actuation speed, which combined are about 50 milliseconds. In the experiment, at most 3 tuning cycles are needed to converge and therefore the total additional tuning time is in the range of 50-150ms.
Dual-wavelength source based optical circuit switching using optimal tuning greatly reduces the power excursions during wavelength reconfiguration for different channel loadings as shown in Fig. 4. Maximum absolute power excursions are reduced to less than 0.3 dB for all channel configurations, with different numbers of new channels (1 channel to 37 channels) added at a time. Blind tuning leads to power excursions under ± 1 dB in most cases but still outperforms single-wavelength source based optical circuit switching. In contrast, using single-wavelength sources, the leftmost wavelength assignment algorithm leads to ± 6 dB power excursions, and therefore requires multiple slow power adjustments to ensure the performance. The centermost wavelength assignment has the most stable performance but still leads to more than ± 3 dB power excursion in the worst case as illustrated in Fig. 4(d). Note that although the single-wavelength random assignment leads to smaller power excursions than other single-wavelength source wavelength assignment algorithms, the random assignment is based on a specific random seed and using another random seed might give a different response and therefore does not capture the worst case response that may occur in a random scenario.
Secondly, we measure the CDF of power excursions using dual-wavelength sources with optimal tuning under four different channel loading plans. For each channel loading plan, 100 random channel configurations are generated and measured. For each channel configuration, 30 values of the absolute maximum power excursions are measured by adding 1-30 new dual-wavelength channels at a time respectively. A channel loading plan includes a specific distribution of the numbers of active channels, the active channel wavelength locations, and the channel hop counts. Figure 5 shows the CDF of four channel loading plans: (i) short-hop sparse normal; (ii) long-hop sparse normal; (iii) short-hop sparse uniform; and (iv) short-hop dense normal. For the short-hop distribution, the average hop count for all measured channels is 1.4, whereas the average hop count is 2.6 in the long-hop distribution. For the sparse distribution, the numbers of active channels range from 5 to 10, while those in the dense distribution range from 20 to 30. In the uniform distribution, the wavelength locations of the active channels are uniformly distributed over the 90-channel spectrum. On the other hand, a Gaussian distribution is used in the normal distribution.
Fig. 5 CDF of power excursions with different channel loading plans. Increased numbers of active channels reduce the power excursions, and the normal distribution outperforms the uniform distribution.
The CDF results of four different channel loading plans are shown in Fig. 5. The third rule of power excursions discussed in Section 2 shows that the power excursions will decrease with the increasing numbers of active channels, which is experimentally verified here such that the ‘short-hop dense normal’ loading plan (magenta cross) outperforms all other loading plans. Its worst absolute maximum power excursion is 0.8 dB and the possibility of the power excursions less than or equal to 0.5 dB is 96.1%. Moreover, the normal distribution (blue triangle) outperforms the uniform distribution (yellow diamond) by 18% at 0.5 dB power excursion margin (89.4% vs. 71.4%), but both distributions give a good performance at 1.5 dB margin (100% vs. 99.0%). The performance differs because wavelengths of active channels are closer to each other in the normal distribution. Therefore, these active channels have a similar gain response, which can be minimized altogether by dual-wavelength sources. In contrast, active channel gains are comparatively different in the uniform distribution, making the power excursion mitigation more difficult. Overall, the performance of different strategies depends on the ripple function of each EDFA.
We also evaluate the dual-wavelength source for power excursion mitigation with different average channel hop counts. The performance of the short-hop distribution (blue triangle) is similar to that of the long-hop distribution (red inverted triangle), although larger power excursions are expected to occur with a cascading hop count. The reason is that the range of the number of active channels (i.e. the total number of channels in the entire system) are kept the same (i.e. 5 to 10) for comparison purposes, and therefore the high average hop count leads to increasing numbers of average channels on each physical link. According to the previous results, the increasing numbers of channels can lead to decreasing power excursions, which as a result cancel out the growing power excursions due to a cascading hop count.
4.3 The combination of dual-wavelength sources and single-wavelength sources
Dual-wavelength sources efficiently reduce the power excursions, with the maximum power excursion less than or equal to ± 1.5 dB 99% of the time and ± 0.5 dB 71.4% of the time for the network considered here, but using multiple dual-switching channels may quickly use up all available wavelengths and reduce the spectral efficiency in a heavily loaded system. One potential solution is time-division multiplexing (TDM), distributing another pair with the same two wavelengths but at disjoint time slots [16]. However, this approach requires a precise global clock to synchronize all laser sources. Fortunately, both the previous mathematical analysis and experimental results have shown that the power excursions decrease with the increasing numbers of channels. This indicates that the dual-wavelength source is beneficial for a ROADM system with a small number of channels, but when the system is heavily loaded, single-wavelength sources can potentially be used with little impact. Furthermore, dual-wavelength sources can be tuned down, removing one of the two sources gradually over time to free up the channel, once sufficient loading is achieved. Therefore, it is critical to determine the minimum number of channels at which dual wavelength sources are no longer needed.
In this section, we investigate the power excursion relation between the number of active channels and the number of new single-wavelength channels. The metric used here is (N:M)margin. N is the number of active channels, and M is the maximum number of new channels added or switched without leading to a maximum power excursion beyond the specified margin, which is set to ± 1.5 dB in this experiment. Using this metric, we can determine the conditions under which dual-wavelength sources need to be used to avoid large power excursions. 1000 active channel loading configurations are randomly loaded for both 10 active channels and 30 active channels, and power excursions are measured as a function of the numbers of new channels using a single-wavelength source. The experimental results of 10 active channels and 30 active channels are shown in Fig. 6(a) and 6(b). The y-axis represents the measured absolute values of the maximum power excursions and the x-axis is the number of new single-wavelength channels added at a time. Using the centermost wavelength assignment algorithm, 6 new channels maximum can be added into the 10 active channel system (i.e. N = 10, M = 6), and 19 new channels maximum can be added into the 30 active channels system (N = 30, M = 19). We see that the (N:M)margin metric is 1.5 for both cases.
Fig. 6 Power excursions vs. the numbers of new single-wavelength channels. Centermost wavelength assignment gives the smallest maximum excursions. (a) 10 active channels; (b) 30 active channels.
By quantifying the metric, we can efficiently implement wavelength reconfiguration with the combination of dual-wavelength sources and single-wavelength sources. Assuming N active channels are already in the system and M new channels will be added or switched during a given period of time, and the system design target is to keep the power excursions less than the ± 1.5 dB margin. According to the metric, if M is less than ⌊N/1.5⌋, all M new channels can be added using single-wavelength sources, without the need of additional slow power adjustments or compensation. And if M is greater than ⌊N/1.5⌋, ⌊N/1.5⌋ channels can be added with single-wavelength sources after M-⌊N/1.5⌋ dual-wavelength source channels are added. Note that this is an ideal situation which assumes that dual-wavelength sources do not introduce additional power excursions. We have shown that dual-wavelength sources still lead to power excursions less than or equal to ± 0.5 dB 71.4% of the time. Therefore, a lower margin should be used, and the maximum number of allowed new single-wavelength channels should be less than ⌊N/1.5⌋ depending on the system design and margins.
5. Conclusion
Using dual-wavelength source based optical circuit switching and wavelength reconfiguration, power excursions are mitigated in a 90-channel 5-ROADM transmission system. A 100 Gb/s PM-QPSK transmission signal is used to evaluate dual-wavelength source performance in a multi-hop coherent system, with less than ± 0.2 dB power excursions. We also investigated the effectiveness of dual-wavelength sources with different channel loadings and the bisection wavelength assignment algorithm. Four different channel loadings are evaluated using both blind tuning and optimal tuning, giving the power excursions of less than 1 dB and 0.3 dB respectively, compared with 6 dB maximum power excursions using single-wavelength sources. The CDF of channel power excursions with optimal tuning is generated from measurements of over 100 random channel loadings for four different channel loading plans. Maximum power excursions are within ± 0.5 dB 71.4% of the time and ± 1.5 dB 99% of the time. Furthermore, wavelength reconfiguration with the combination of dual-wavelength sources and single-wavelength sources has been investigated. For the 5 ROADM system studied here, it was found that with N active channels, ⌊N/1.5⌋ new channels can be added using a single-wavelength source. Likewise, when the number of active channels increases such that the ⌊N/1.5⌋ condition is exceeded, then dual-wavelength sources can be migrated to single-wavelength sources, recovering the system spectral efficiency.
Funding
National Science Foundation (EEC-0812072); National Science Foundation (1601784; Department of Energy DE-SC0015867).
Acknowledgements
The authors would thank for A.P. Anthur, Z. Qu, L. Barry, and M. Cvijetic for their technical insight and informative discussion.
References and links
1. “Cisco Virtual Network Index” [Online]. http://www.cisco.com/c/en/us/solutions/collateral/service-provider/visual-networking-index-vni/vni-hyperconnectivity-wp.html
2. “Cisco Global Cloud Index: Forecast and Methodology 2013–2018,” Cisco White Paper [Online]. http://www.cisco.com/c/dam/en/us/solutions/collateral/service-provider/global-cloud-index-gci/white-paper-c11-738085.pdf#_ftnref1
3. A. Mahimkar, A. Chiu, R. Doverspike, M. D. Feuer, P. Magil, E. Mavrogiorigis, J. Pastor, S. L. Woodward, and J. Yates, “Bandwidth on demand for inter-data center communication,” Proceedings of the 10th ACM Workshop on Hot Topics in Networks. ACM, 2011.
4. D. C. Kilper, K. Bergman, V. W. S. Chan, I. Monga, G. Porter, and K. Rauschenbach, “Optical Networks Come of Age,” Opt. Photonics News 25(9), 50–57 (2014).
5. I. Tomkos, S. Azodolmolky, J. Sole-Pareta, D. Careglio, and E. Palkopoulou, “A tutorial on the flexible optical networking paradigm: State of the art, trends, and research challenges,” Proc. IEEE 102(9), 1317 (2014).
6. “Introduction to EDFA Technology,” Finisar White Paper, June 2009. [Online]. https://www.finisar.com/sites/default/files/resources/Introduction%20to%20EDFA%20technology.pdf
7. C. Tian and S. Kinoshita, “Analysis and control of transient dynamics of EDFA pumped by 1480-and 980-nm lasers,” J. Lightwave Technol. 21(8), 1728 (2003).
8. L. Rapp, “Transient performance of erbium-doped fiber amplifiers using a new feedforward control taking into account wavelength dependence,” J. Opt. Commun. Netw. 28(2), 82 (2007).
9. L. E. Nelson, G. Zhang, N. Padi, C. Skolnick, K. Benson, T. Kaylor, S. Iwamatsu, R. Inderst, F. Marques, D. Fonseca, M. Du, T. Downs, T. Scherer, C. Cole, Y. Zhou, P. Brooks, and A. Schubert, “SDN-Controlled 400GbE end-to-end service using a CFP8 client over a deployed, commercial flexible ROADM system,” Optical Fiber Communication Conference Postdeadline, Th5A. 1, 2017.
10. D. C. Kilper, C. A. White, and S. Chandrasekhar, “Control of Channel Power Instabilities in Constant-Gain Amplified Transparent Networks Using Scalable Mesh Scheduling,” J. Lightw. Tech 26(1), 108 (2008).
11. Y. Huang, W. Samoud, C. L. Gutterman, C. Ware, M. Lourdiane, G. Zussman, P. Samadi, and K. Bergman, “A Machine Learning Approach for Dynamic Optical Channel Add/Drop Strategies that Minimize EDFA Power Excursions,” European Conference on Optical Communication, 2016.
12. Y. Huang, C. L. Gutterman, P. Samadi, P. B. Cho, W. Samoud, C. Ware, M. Lourdiane, G. Zussman, and K. Bergman, “Dynamic mitigation of EDFA power excursions with machine learning,” Opt. Express 25(4), 2245–2258 (2017).
13. U. Moura, M. Garrich, H. Carvalho, M. Svolenski, A. Andrade, F. Margarido, A. C. Cesar, E. Conforti, and J. Oliveira, “SDN-enabled EDFA Gain Adjustment Cognitive Methodology for Dynamic Optical Networks,” European Conference on Optical Communication, 2015.
14. K. Ishii, J. Kurumida, and S. Namiki, “Experimental Investigation of Gain Offset Behavior of Feedforward-Controlled WDM AGC EDFA Under Various Dynamic Wavelength Allocations,” IEEE Photonics J. 8(1), 7901713 (2016).
15. K. Ishii, J. Kurumida, and S. Namiki, “Wavelength assignment dependency of AGC EDFA gain offset under dynamic optical circuit switching,” in Optical Fiber Communication Conference, W3E.4, 2014.
16. A. S. Ahsan, C. Browning, M. S. Wang, K. Bergman, D. C. Kilper, and L. P. Barry, “Excursion-Free Dynamic Wavelength Switching in Amplified Optical Networks,” JOCN 7(9), 2015.
17. S. Zhu, W. Mo, D. C. Kilper, A. P. Anthur, and L. Barry, “Dual Laser Switching for Dynamic Wavelength Operation in Amplified Optical Transmission,” Optical Fiber Communication Conf. (OFC), Th2A.44 2017.
18. R. Maher, D. S. Millar, S. J. Savory, and B. C. Thomsen, “Widely Tunable Burst Mode Digital Coherent Receiver With Fast Reconfiguration Time for 112 Gb/s DP-QPSK WDM Networks,” J. Lightw. Tech 30(24), 3924 (2012).
19. J.A. O’Dowd, K. Shi, A. J. Walsh, V. M. Bessler, F. Smyth, T. N. Huynh, L. P. Barry, and A.D. Ellis, “Time Resolved Bit Error Rate Analysis of a Fast Switching Tunable Laser for Use in Optically Switched Networks,” JOCN 4(9), A77 (2012).
20. J. Junio, D. C. Kilper, and V.W.S. Chan, “Channel Power Excursions from Single-Step Channel Provisioning,” JOCN 4(9), A1 (2012).
21. J. Zyskind and A. Strivastava, Optical Amplifiers for Advanced Communications Systems and Network (Academic, 2010), Chapter 8.
22. T. J. Xia, G. A. Wellbrock, Y. Huang, E. Ip, M. Huang, Y. Shao, T. Wang, Y. Aono, T. Tajima, S. Murakami, and M. Cvijetic, “Field experiment with mixed line-rate transmission (112-Gb/s, 450-Gb/s, and 1.15-Tb/s) over 3,560 km of installed fiber using filterless coherent receiver and EDFAs only,” Optical Fiber Communication Conference and Exposition, pp. 1–3, 2011.
