Phyisical Impairment Constrained Operation
3.2 Joint Power Level Tuning and Routing
Let us consider the second scenario, where we try to fight against impairments by increasing the power level of certain signals. In this case the BER drops as the level of the signal grows, however, only until a certain threshold. When the total of channel powers in a certain link achieves a threshold, the transmission impairments will rapidly escalate due to non-linear effects!
To better explain the problem, let us consider Figure 3.1 again, however, this time assuming no grooming capability at all. Considering that link a-b is used by the largest number of demands it will have the highest total power level. If we want to improve the quality of the signal between any of the s-dnode-pairs, we have to increase the power. However, as we increase the power of certain channels, the total power grows as well, that leads to risen nonlinear impairments. To avoid this problem we propose two solutions.
First, routing some of the demands to longer paths (solid instead of dashed path in Figure 3.1) decreases the total power of critical links. This will need higher power for that wavelength-path, however, as long as it does not use any critical link it is no problem, it may have higher power level.
Second, we propose using different power levels for different wavelength channels even within a single link. Although the optical equipment allows this uneven tuning, to our knowledge it has not been used so far. Nowadays in nearly all reconfigurable optical add-drop multiplexers (ROADM) the signal power can be tuned this way by the control plane via variable optical attenuators (VOA).
The proposed methods can be used in existing WDM optical networks wherever the nodes support signal power tuning.
3.2.1 Technology Background of PICR
New technologies aim to reduce operational expenditure (OPEX). Reconfigurable optical add/drop multiplexers (ROADM) provide remote configuration capability, including capacity and power tun-ing without manual intervention for a wide range of wavelengths.
Most of the operators, to optimise the performance of their networks require monitoring and wavelength control. Thus, additional management functions, that allow power measurement and other per wavelength settings are included in most of the commercial products [19], [69].
Nowadays in nearly all types of ROADMs signal power can be tuned with variable optical
0.25 0.35 0.45 0.55 0.65
Experiment 2 - Average of optical hops
Expansion
Figure 3.4: Average per-demand hopcount: Dependence on the network scale (’Expansion’) and grooming capacity(’Port number’).
attenuators (VOA) through the management system. This, along with the Routing and Wavelength Assignment (RWA), enables fully reconfigurable networks.
The details of the configuration, design, and optimization of optical networks have been inten-sively investigated so far, see e.g. [86]. However, only few papers focused on the physical parameters ofRWAin case of fully reconfigurable optical networks [40], [C124]. To our knowledge, none of these examined the idea of handling physical impairments by joint power level tuning and routing.
In metroWDM networks signal power of optical channels is determined by Cross-Phase Mod-ulation (XPM) and Raman scattering and not by the Brillouin threshold. This means that the maximum of total power inserted into a fiber is limited, not the power used to transfer single de-mands. Thus, it is suitable to increase the power of some channels up to the Brillouin threshold while other channel powers are tuned down to fulfill the XPM and Raman scattering constraints. This idea allows the use of Physical Impairment Constrained Routing (PICR) for lightpath configuration [C112].
In Fig. 3.7 we have two wavelengths: φ1 andφ2. InCase Awe do not applyPICR. Here, due to physical constraints, nodeAcan only reach nodeC in all-optical way. If there is a demand between node A and D its path can only be established with electric signal regeneration either in node B or in node C. Case B shows the same situation when the proposed PICR approach is used. Here the signal power of φ2 is increased to fulfill the Optical Signal-To-Noise Ratio (OSNR) requirement at node D. This can be done only, if the total power load is affordable on each link. Therefore, the power level of φ1 has to be decreased. This way it is possible to establish an all-optical connection betweenA and D.
3.2.2 The MILP Formulation of the Problem
In [P4] and [C112] we propose a new method for finding the global optimum of the wavelength-path system configuration by simultaneously tuning the power levels for each wavelength wavelength-path and routing these wavelength paths in order to minimize the effects of non-linear impairments while maintaining the sufficient signal level for required end-to-end signal quality in terms of BER. This method also minimizes the use of network resources within the constraints of end-to-end BER for each wavelength path. These approaches are based on ILP (Integer Linear Programming) and on heuristics. If there exists a global optimum the ILP algorithm will find it, for any network
0.25 0.35 0.45 0.55 0.65 20
24 28
32 36
80 0.001
0.002 0.002 0.003 0.004 0.005 0.006 Regeneration/
Demand
Experiment 2 - Average of regenerations due to physical impairments
Expansion Port number
Regeneration/
Demand
0.000 0.001 0.002 0.003 0.004 0.004 0.005 0.006
Figure 3.5: Average per-demand number of those regenerations that were required explicitly because of physical impairments: Dependence on the network scale (’Expansion’) and grooming capacity (’Port number’).
RWA 1 1,2 1,4 1,6 1,8 2 8
0 10 20 30 40 50 60 70 80
0,5 scale 0,75 scale 1 scale 1,25 scale 1,5 scale
Routed demand
n-factor
Figure 3.6: Maximum number of routed demands versus the n-factor for different scale parameters.
Figure 3.7: Base idea of PICR
topology, physical constraint and demand set. In [C112] we generalize the results for two-layer grooming-capable networks as well.
Here we provide the ILP (or rather the MILP: Mixed Integer Linear Programming) formulation of the problem for single layer wavelength routed single-hop networks.
We assume the network to be modeled as a wavelength graph that consists of vertices V that correspond to the ports at different wavelengths of the switches that are interconnected within switches by edges from set Esw. The edges that are used to interconnect ports of different switches are from setE\Esw. These edges from the setEswbelong to different physical links (fibers)pl. All thepls make the set of physical linksP L. Considering all the edges (i, j) belonging to∀pl∈P Lwe will have the set E\Esw.
The set of demands o is denoted as O. A single demand o will be routed using a single, end-to-end wavelength channel. The signal will be amplified all optically to maintain its power level, however, no 3R regeneration is assumed.
Constants:
• Ppl[mWmax ]is the upper limit of total power in physical link (fiber)plexpressed inmW. (P[dBm]= 10 log10P[mW], i.e., 1 mW corresponds to 0 dBm.) The total power level in a SMF fiber must never exceed 100mW (Ppl[mWmax ]<100mW,Ppl[dBm]max <20dBm) and has typical value of over 3 mW (4.77 dBm). Depending on the network parameters (including the number of wavelengths used) the value of Ppl[mWmax ] is set between these limits.
• lij is the length of a physical link between nodes i and j expressed in [km]. If we want to model physical impairments induced by a node, we assume that the impairments induced by the switch are equivalent to that experienced while sending the signal along a certain length of a fiber, (e.g., 90 km). The simplest way to include physical impairments induced by a node is to add this length to fibers.
• Lcis a linear factor between the distance the signal has to reach and the input power required for reaching that distance with acceptable BER. It is expressed in [km/mW]. As the input power level is increased, the signal quality will improve, however, after a certain level the nonlinear impairments will significantly deteriorate the signal quality. Here we have assumed Lc= 1000 km/mW for a singleλ.
• α, 0 < α < 1 is a tuning parameter that weights the optimisation objectives. It can prefer either the minimal routing cost (largerα) or the minimal total power used (smaller α).
• n-factor is the value of maximal relative deviation from average per channel power level for a singleλ(wavelength). If there are|λ|wavelengths per channel the average per channel power level will be Ppl[mWmax ]/|λ|. If the n-factor has value of 1 all the channels must have this equal power level. If it has value of e.g. 1.5 it means that a channel can have power level by 50 % higher, however, then it may happen that the other λ(wavelength) channels must decrease their power level in order not to exceed the total per fiber power level.
• so and to are the source and the target (destination) respectively of demando∈O.
Variables:
• p0,0 ≤ p0 ≤ n/|λ|,∀o ∈ O is the input power of demand o expressed in mW and then normalised by Ppl[mWmax ]
• p0ij,0≤poij ≤po,∀(i, j)∈E,∀o∈O is the power of demand oon link (edge) (i, j) normalised by Ppl[mWmax ]
• yij0 ∈ {0,1},∀(i, j)∈E,∀o∈O is the binary indicator variable having value of 1 if demando
• Constraints 3.2 and 3.3 are the flow conservation (real) variables for the power and for the flow indicator (0-1) respectively.
• Constraint 3.4 ensures, that whenever the power of demand o on edge (i, j) exceeds 0, than edge is considered used via the 0-1 indicator variableyijo.
• Constraint 3.5 guarantees, that the total power limit of a physical link (fiber) is not exceeded.
• Constraint 3.6 ensures that a single wavelength of a fiber can be either not used or used by a single demand only.
• Constraint 3.7 ensures that the reach of demand ocannot exceed the distance determined by its input power. Violating this constraint would lead to signal quality deterioration expressed as BER or Q-factor.
3.2.3 Results
We have evaluated the proposed ILP based approach for the same network, shown in Figure 2.36(a).
The results are presented in Figures 3.6 and 3.8. In both figures we evaluate the number of demands that can be simultaneously routed for certainn-factors. n=1 means, that all channels have the same power level, i.e., the maximum power allowed in a fiber is divided by the number of wavelength channels.
Figure 3.6 shows, that as we allow more and more diverse power levels, the amount of demands that can be routed grows significantly, e.g., from 18 to 70 for scale factor 1 while the n-factor grows
RWA 1 1,25 1,5 1,75 2 0
10 20 30 40 50 60 70 80
70
13 31 26 54
Routed Demands
n-factor 2 wavelengths
3 wavelengths 4 wavelengths 6 wavelengths 8 wavelengths
Figure 3.8: Maximum number of routed demands versus the n-factor for different number of wave-lengths.
from 1 to 1.4. This shows the great performance of allowing this minor difference in power levels! If we consider the scale factors, it can be seen that for value of 0.5 where physical impairments do not impact the number of routed demands at all, there is no difference - no need to use different power levels. However, for scale factors larger than 1, where the physical impairments become significant allowing different power levels results a tremendous growth in the number of routed demands. The first value on the horizontal axis marked as RWA is the reference method, where all the power levels are exactly the same.
Figure 3.8 shows the dependence of the number of routed demands on the n-factor as the number of wavelength channels is increased in the links of the WDM network. It can be seen that as the number of wavelengths grows, the proposed approach of allowing different power levels has an incredible positive impact: Instead of a single demand, up to 70 demands can be routed in a system with 8 wavelengths as the n-factor grows from 1 to 2!
3.2.4 Increased Throughput through Proposed PICR-Aware Methods
In Section 3.1 and in Section 3.2 we have proposed two scenarios for Physical Impairment Con-strained Routing (PICR). First, in Section 3.1, using the grooming capability present in certain nodes of the network to perform O/E/O signal regeneration. Second, in Section 3.2, to tune the signal power of certain wavelength-paths to different levels depending on the destination of that wavelength path, as well as on the current power budget of the links along that path. Both schemes were used simultaneously with routing over optical C/D WDM networks. Our simulations have shown that using the proposed schemes presented in Sections 3.1 and 3.2 the ratio of demands routed can be significantly increased compared to the cases when all demands are blocked that could have been routed by conventional methods, however, their signal quality was poor due to the physical impairments.