Login| Sign Up| Help| Contact|

Patent Searching and Data

Document Type and Number:
WIPO Patent Application WO/2002/033517
Kind Code:
A closed loop amplified ring for metropolitan (Fig. 3) area DWDM networks. The ring includes a reference node (104) where channels that originate at the reference node are added to the ring at a reference power level and where the reference power level is used to set power levels of all other channels that are added to the ring.

Application Number:
Publication Date:
April 25, 2002
Filing Date:
September 27, 2001
Export Citation:
Click for automatic bibliography generation   Help
International Classes:
H04J14/02; (IPC1-7): G06F/
Foreign References:
Attorney, Agent or Firm:
Kirkland, Mark D. (Suite 500 Redwood City, CA, US)
Download PDF:
CLOSED LOOP AMPLIFIED RINGS INCLUDING LOSSY SPANS The present invention is directed to network communications systems and more specifically to a system and method to a stable closed ring-based wavelength division multiplexing system.

Background In a simple form, a communications network includes a plurality of stations coupled by a transmission media (e. g., cable wire or optical fiber) over which the stations communicate. Examples of communications networks include telecommunication systems, cable television systems and local area or other computer networks. The communication infrastructure (e. g., the transmission media such as optical fiber, cable wire etc.) for a communications network includes a physical configuration that is referred to as a topology. A ring topology, for example, connects network nodes in a loop or ring.

Information is transferred from a source node to a next node, and so on, around the ring to reach a destination node. A ring topology has the advantage of minimizing the amount of transmission media that must be used to connect the nodes in the network. However, the amount of information that can be transmitted (i. e., the bandwidth) is limited in a ring topology.

In contrast to a ring topology, a star topology connects branch nodes to a central node in a spoke-like fashion. Information is transmitted from one branch node to another via the central node. A star topology has the advantage of having a central node that can be used to link to another communications system. Further, a branch node that is connected to the central node can use the full bandwidth of the connection between a respective node and the central node. That is, the branch node does not need to share any bandwidth between itself and other nodes when communicating with the central node.

The star topology advantageously can provide the full bandwidth between a branch node and a central node. However, the communications infrastructure required to support a star topology is significantly greater than that for a ring topology.

As described above, in a ring topology a single communication channel is shared between the nodes connected on the ring. In an effort to increase the bandwidth of a conventional ring configuration, two mechanisms have been developed for proportioning the networks bandwidth. These two mechanisms are time division multiplexing (TDM) and wavelength division multiplexing (WDM).

In a TDM system, the transmission bandwidth of the ring's single communication channel is broken up into intervals or slots of time. In a given time interval a node is given that single channel's full bandwidth. Where there are n nodes, there are-time n slots. Thus, each of the n nodes may transmit information-of the time. Each node is enabled to send up to the full bandwidth of data in its respective time slot. The length of time associated with a time slot may necessitate information being split up into multiple transmissions or packets that are combined to form a complete transmission at a destination node. A disadvantage of a TDM system is that the nodes still may share only a fraction of the total available bandwidth of the communications network. The more nodes present in the communications network, the smaller the amount of time allotted to each node.

Instead of dividing a network's bandwidth into time slots, a WDM system divides a network's bandwidth into signal channels, where each signal channel is assigned a particular channel wavelength. This allows multiple signals (each a different wavelength) to be carried on the same transmission media. For example, multiple optical signal channels can be used by a fiber optic cable to transmit multiple signals on the same cable. Each signal channel operates at the network's full bandwidth. Thus, a node can use the full bandwidth of the network by sending information on one of these signal channels.

The signals are multiplexed in a WDM system at a transmitting end and transmitted to a receiving end where they are demultiplexed into individual signals. In conventional systems, the transmitting and receiving ends must be tuned to the same wavelengths to be able to communicate. That is, the transmitting and receiving ends use a device such as an add/drop multiplexor to transmit/receive a fixed signal channel. In the case of fiber optic cable, an add/drop multiplexor is used at the transmitting and receiving ends to generate a fixed wavelength (e. g., using lasers) signal and to receive a fixed wavelength signal. Conventional systems can have as many as 16 to 40 signal channels.

Metropolitan area networks in a star or ring configuration typically consist of core and edge nodes. The core nodes connect to a long haul backbone and direct the metro traffic going in and out of the edge nodes. The edge nodes accommodate all the data and voice services demanded by an access network connected to an edge node that serves business and residential customers.

Fiber optic rings utilizing dense wavelength division multiplexing (DWDM) are essential for meeting the large bandwidth demands of businesses and residencies in the access, metropolitan area, and regional marketplaces. In a typical configuration, a ring is deployed with multiple nodes dotting the perimeter where bandwidth can be added and dropped from the ring. The add-drop functionality in DWDM systems is implemented in the optical domain by extracting from or added to the ring a full wavelength using optical filters known as optical add-drop multiplexers.

Long fiber lengths create loss, and the filters used to add and drop channels also create loss. A network that can scale with the ring perimeter and the node number must use optical amplifiers along the ring to periodically boost power levels. In general, the scalability of the network, as it relates to optical power, is only limited by the maximum number of amplifiers that can be placed along the ring.

Fiber optic rings can be of various types. Examples include broken and closed rings. In a broken ring topology, optical signals never go around the ring more than once.

This is advantageous with respect to optical amplifiers, because amplified spontaneous emission (ASE) cannot recirculate multiple times around the ring. A star-over-ring architecture, in which all nodes communicate with a central core on the ring, but do not communicate directly with each other, is consistent with the broken ring topology.

Large bandwidth demand between two nodes can require a dedicated wavelength connection. In this case, directly connecting two nodes is often more efficient and cost effective than connecting those nodes indirectly through a central core. This creates a sort of decentralized ring network, and requires a closed ring topology. A closed ring without optical amplifiers is simple to implement. Power levels around the ring need not follow any particular pattern ; the only constraint on power levels is that the incoherent channel crosstalk at a drop site-which is generated by the other channel powers on the ring-be kept below some minimum value. A typical requirement for incoherent crosstalk is that it must be 15 dB below the desired drop channel power at the receiver.

The maximum allowed power for unselected channels at the input to the drop filter is easily derived from the drop filter's isolation specifications. High degrees of isolation, as are found in dielectric filters, ease the power level requirements, and a typical implementation therefore involves coupling the maximum power from the transmitter onto the ring. This allows for the greatest distance between the originating add site and the farthest drop site, and the greatest number of intermediate nodes between them. In a protected ring environment, there are two paths of communication between any two nodes-often shown schematically with a clockwise and counterclockwise path.

An unamplified ring is easy to implement, but does not scale with respect to the perimeter of the fiber ring, or the number of nodes in the system. As a result, this architecture has limited applications. The introduction of optical amplifiers onto the ring enables system scalability, and is therefore vital for any kind of extended distance or large-node applications. The presence of amplifiers on a closed ring, however, introduces a complication. Optical amplifiers generate ASE (noise) over a wide band, and most of the power is emitted at wavelengths that are not terminated at the add-drop multiplexers. The ASE will therefore circulate around the ring multiple times. The extent to which it recirculates depends on both the ring loss and the total amplifier gain (gain from all amplifiers combined) around the ring. When the total amplifier gain is greater than the loss introduced by the fiber and the various passive filtering components at the add-drop nodes, the fiber ring can become a fiber ring laser. This is clearly an undesirable situation as it will bring down all the amplified traffic on the ring, and potentially the unamplified traffic also. What is desirable is a system that includes the benefits of a closed ring topology that provides scalability and stability for DWDM applications.

Summary In one aspect the invention provides a closed loop amplified ring for metropolitan area DWDM networks. The ring includes a reference node where channels that originate at the reference node are added to the ring at a reference power level and where the reference power level is used to set power levels of all other channels that are added to the ring. The ring includes one or more edge nodes, links coupling each node in the network to another to form the ring and a lossy span associated with one or more links. The lossy span includes one or more amplifiers and an attenuator where the gain of the amplifier at the end of the lossy span is less than the span loss on the associated link including the loss on the link and the attenuation introduced by the attenuator. The ring includes means at each node that, for each channel on the ring that traverses the lossy span, adds a predefined amount of extra power relative to the reference power.

Aspects of the invention can include one or more of the following features. The closed loop amplified ring wherein the amplifier at the end of the lossy span is connected to an ingress of the reference node.

The details of one or more implementations of the invention are set forth in the accompanying drawings and the description below. Other features, objects, and advantages of the invention will be apparent from the description and drawings, and from the claims.

DESCRIPTION OF DRAWINGS FIG. 1 a shows a closed ring topology fiber optic ring.

FIG. lb shows a reference node.

FIG. 2 shows a closed ring topology including plural amplifiers.

FIG. 3 shows a closed ring topology including three amplifiers linked by two spans and a lossy span.

FIG. 4 shows a core design for a mixed star-over-ring and mesh-over-ring architectures.

FIG. 5 shows a fully reconfigurable mixed star-over-ring and mesh-over-ring architecture.

FIG. 6 shows an alternative a core design for a mixed star-over-ring and mesh- over-ring architectures.

FIG. 7 shows an alternative core similar to that shown in FIG. 6, but with the signal channels switched.

Detailed Description Referring now to FIG. la, a closed loop ring 100 is shown. The closed loop ring includes plural edge nodes 102b-d and a central node 104. Nodes are connected by media 106. A single direction of traffic flow is shown. However, a protection path can be added to support bi-directional communication around the closed loop ring. Each path (loop) can be treated separately, and as such, only a single path having traffic flowing in a clockwise direction on the page is discussed below.

Associated with the path is a lossy span 110. The lossy span 110 includes an attenuator 112 and an amplifier 114. All optical signals that traverse the lossy span pass through both the attenuator 112 and amplifier 114. The net gain from all amplifiers on the closed loop ring 100 is set to be less than the net loss around the ring from the fiber (media) and other passive components (e. g., attenuator 112). In this implementation, the net gain of the amplifier 114 is set to be less than the net loss around the ring plus the attenuation of the attenuator 112. The operation of the attenuator and amplifier is described in greater detail below.

Channel power levels Central node 104 is a reference node in the system, for the purpose of setting power levels. Any of the nodes could be chosen; in the practical deployment of a network, it is probably sensible to define the node that connects to the backbone network as the reference node. Power levels are set there initially, and signals originating at other nodes are matched to these reference levels as will be described below.

The power levels for any signals sharing a common amplifier on the ring 100 should ideally be equal. For example, most channels originating at node 102c (except those destined for node 102d) and all channels originating from node 102d should be power matched to each other. This is because they share a common amplifier 114. The only requirement on other channels that do not share a common amplifier is that they don't violate the incoherent crosstalk specifications of other channels dropped at a respective node. This is the same condition discussed above for the unamplified closed loop ring.

In general, signals passing through the reference node 104 should have equal power to the signals originating at the reference node 104. However, the power level of an express channel (a signal that passes through the reference node) is set at its originating node. This implies that the gain around the ring must be equal to the loss-at least as far as the active channels are concerned. Equalizing channel gain and loss around the ring is different from equalizing cavity gain and loss. The latter is a hard requirement that must be adhered to so as to prevent the ring from oscillating. The former is a condition that can be met-even when cavity gain is less than cavity loss-by compensating the excess cavity loss with extra power from the laser transmitter at an add- drop node.

Accordingly, if the loss for the lossy span 110 is S+t and then gain for the amplifier 114 is set to S. In addition, signals originating at nodes that will traverse the lossy span 110 have their signal levels increased by the delta amount (A). The delta (A) amount of loss can be set using the attenuator 112. Attenuator 112 can be a variable optical attenuator. In this configuration, the channel gain around the ring is equal to the channel loss, but for which the cavity gain is less than the cavity loss. As discussed above, the excess cavity loss is offset by bumping up the power level of the add channel for each affected channel by a specified amount, the delta (A). The value of the delta should not be too high, because signals passing through common amplifiers around the ring will not be power matched. Thus, the power level for the channels added at an affected node is slightly higher than the power dropped. As a result, the power level of this added signal will differ slightly from a signal that either originates at the reference node or terminates before the amplifier, or both. The effect of this power mismatch needs to be considered in a real network, where there is more than one in-line amplifier.

When the mismatched signals enter another amplifier in the network, that amplifier must be able to tolerate the power difference without degrading the performance of any of the channels in the system.

Referring now to FIG. lb, the reference node 104 is shown with 6 incoming channels that are attenuated (by attenuator 112) and then amplified (by amplifier 114).

Three of the incoming channels are dropped (the three terminated at the reference node from each of nodes 102b-d, respectively) at add drop multiplexors 120-1,120-2 and 120- 3, and their corresponding add channels serve as the reference powers for the node (and the entire ring). The other three channels (carrying signals from nodes 102d to 102b and 102c as well as signals from node 102c to node 102b) pass through the reference node and are referred to as express channels. The output power of the amplifier 114 is set, so that at the output of the node, the express channels and the reference channels are power matched. Note that it is precisely the power matching condition at the reference node that forces the express channels to be added onto the ring with extra power. Because the amplifier 112 is operating at a constant gain that is less than the span loss (S+ A) before it, express and reference channel power matching at the reference node can only be achieved by boosting the express channel powers from the add site (for this node, the sources of origin for the express channels, nodes 102d and 102c), so as to compensate the excess span loss. In Figure lb, PIR, P2R, and P3R are the reference channels; P4E, PsE, and P6E are the express channels.

In a fully connected network, at each other node, a similar number of channels are dropped and added, then coupled with a similar number of express channels. At these other nodes, attenuators and amplifiers may or may not be present at the ingress or egress for a given node as discussed below.

Plural lossy links and uneven distribution of Nodes The proposed closed loop ring may have nodes that are uniformly spaced or unevenly spaced and as such may include one or more lossy spans. The implementation shown in FIG. la includes one amplifier in a single lossy span. One of ordinary skill in the art will recognize that a uniformly spaced distribution of nodes will never require a single amplifier. If any amplification is required, two amplifiers will be present in the network. The present invention has been discussed with reference to a fully connected mesh (all nodes connected to each other). A mesh that is not fully connected can have some nodes with more loss than other nodes, thus creating an asymmetry analogous to that created by different fiber lengths. While an initial network deployment may not demand full connectivity, the proposed network topology can be expanded to support full connectivity, and as such is described herein with reference to a fully connected mesh.

Referring now to FIG. 2, a closed loop ring 200 having uniformly distributed nodes includes nodes 202b-d and a central node 204. The closed loop ring 200 includes a lossy path 206 having an attenuator 212 and a first amplifier 214. The path between nodes 202c and 202d includes a second amplifier 216. In this configuration, each node in the closed loop ring 200 will always have at least one channel that will ideally be power matched to a channel from every other node on the ring. While every single channel on the ring does not have to conform to this strict power matching requirement (as described above with respect to the power increase of delta for each channel that traverses the lossy span 206), in practice, the channel powers of signals originating at a common node will be approximately same (within the delta).

Referring now to FIG. 3, a multinode closed loop amplified network 300 configuration is shown. In this configuration, at least one amplifier will not fully compensate the span loss (including fiber loss and add-drop filter losses) along one link (a link or span is defined as the connection between two amplifier sites) in the network.

The network 300 includes three amplifiers 302, each amplifier having gain S. The span loss for the so-called lossy span is set at S+ (A), whereas the other links have a loss of S.

Thus, the regular links, in this network, function in a very similar manner as the links in long haul systems, where the loss along each link is compensated exactly by the gain of the amplifiers. In the metro environment, this kind of architecture can be implemented by placing a variable optical attenuator in front of each amplifier, to equalize variations in span loss from node to node. The lossy span, on the other hand, is required to break the symmetry and fulfill the requirement that cavity gain be less than cavity loss.

Reference Node Placement As discussed in the previous section, some channel powers on the ring must be added at higher levels than the power to which they are referenced. In fact, the channels requiring this excess power are exactly those channels that traverse the lossy span.

Traversing the lossy span, in this context, means that the channel passes through the amplifier at the end of the span-the amplifier that does not fully compensate the span loss. Note that the channel does not have to traverse the entire span-it can be added in the middle of it, but it must pass through the amplifier at the end. Likewise, a channel that is dropped off before passing through the end of span amplifier does not have to be added at a higher power level. In the closed ring amplified architecture, the power budget around the ring is set by assuming that any channel exiting an in-line amplifier (including the one after the lossy span) will have a minimum output power. If the output power per channel drops below that number, the performance of certain connections around the ring can degrade. The higher add power is therefore required on channels that traverse the lossy span, but is not required for channels that are dropped off the ring before that link.

The lossy span, in principle, can be placed anywhere along the network ring, and will prevent the ring from lasing. However, if the lossy span is not the connection immediately before or immediately after the reference node, the reference powers-if set properly-will not all be equal. This results from the fact that some channels originating at the reference node will pass through the lossy span, while others will not when the lossy span is placed somewhere in the middle of the network. This is an undesirable situation for a reference node, and is not practical.

In order to keep the reference node channel powers equal, the lossy span should connect a node on the ring with the reference node. The topology of the network allows for several configurations-for example, an amplifier could be placed at the node preceding the ingress to the reference node (at the output of that node), or at the first node after its egress (at the input of that node), or at the reference node itself. The first two configurations have limited use, because they don't allow for the possibility of an unamplified node between the reference node and its adjacent nodes. Also, the reference node power levels will typically be high, and most likely, an amplifier will be required there to match the power levels of pass-through (express) channels (that is, those not originating at the reference node) to the reference channels. Accordingly, a best solution may be to place an amplifier at the ingress of the reference node, and to create the excess loss on the link by using a variable optical attenuator at the amplifier input. The lossy span condition is achieved by operating the amplifier at constant gain and by increasing the VOA attenuation to ensure that the total span loss is greater than that gain.

If the express channels are added at 3 dB higher power than the reference and normal channels, a common amplifier (e. g., Amplifier 114 of FIG. la) must be able to tolerate a 50% increase in input power without system degradation. The assumption is that small variations in power levels do not significantly impact amplifier performance from channel to channel (output power, SNR). In fact, power level variation between channels will be present in any network, due to finite gain flatness of amplifiers, nonuniformity in transmitter powers across a band, and the wavelength dependent loss in passive components such as multiplexers and add-drop filters. The key point is that the network design has to include margin in the power budget to accommodate these variations. As long as the variations are kept small enough, catastrophic network failures will not occur.

Alternative Implementations As described above, a network can initially be built with just the reference node amplifier to start, and no amplifiers in the rest of the network. This configuration (as shown in FIG. 1) allows for future upgrades to the network. If an amplifier needs to be placed in the system at a later time, the reference node amplifier is essential. However, if no upgrade will ever be required, and no other amplifiers are needed in the system, the reference node amplifier can be dispensed with and the signals around the ring do not need power matching. This is just the unamplified closed ring network discussed at the beginning of the document.

Some applications do not demand full wavelength connectivity between every node on the ring. A star over ring architecture is more applicable for these applications, since it more efficiently utilizes the wavelength resources on the ring. In the star over ring architecture each node has a full wavelength connection to a central core, rather than to every other node on the ring. The central core redistributes data around the ring to different nodes. Typically, there will be equipment present at the core that can perform this rearrangement and redistribution ("grooming") of data. Thus, a pair of nodes can communicate with each other at a granularity that is smaller than a full wavelength. This is a great advantage in cost and efficiency if those nodes do not require a full wavelength worth of bandwidth between them. The difference therefore, between the star over ring and the mesh-over-ring architectures, as far as bandwidth is concerned, is that in the star over ring architecture an edge node shares its full bandwidth with all the other nodes on the ring. For example, a ring could have 16 edge nodes, yet only one OC48 wavelength at each node. The average connectivity between nodes would then be OC3. In contrast, in the mesh-over-ring architecture, a wavelength originating at an edge is not shared with the other nodes but is fully dedicated to only one other node on the ring.

Applications can exist which require a mixture of star over ring and mesh-over- ring. In this scenario, some nodes demand a sub-wavelength granularity connection, while others demand a full wavelength connection. The two architectures can be combined, by defining the core as the reference node. At the core, inbound star-over-ring and mesh-over-ring channels can be separated, with the former terminated and reinserted onto the ring-thus serving as the reference channels-and the latter passing through an amplifier at the core. Identical to the situation in the full mesh over ring, the core amplifier can be set so that the express channel powers match the reference channel powers. The same design rules that are used for building the full mesh over ring structure can be used for the mesh over ring channels in the mixed architecture. Excess loss can be generated with an attenuator in front of the core amplifier, and the express channels will be added onto the ring with extra power so as to compensate the excess loss of the lossy span.

A few core/reference node configurations for the mixed architecture are shown in FIGs 4-7. In FIG 4, the star-over-ring channels and the mesh-over-ring channels are separated into different bands. A band-splitting filter 402 at the input of the core separates the two bands, with the star-over-ring passing to a demultiplexer 404 for termination, and the mesh-over-ring passing to the core attenuator 405 and amplifier 406 pair. The star-over-ring channels are reinserted onto the ring through a multiplexer 408, and combined with the mesh-over-ring channels (coming out of the amplifier) with a band-combining filter 410.

Figure 4 shows a particular design where the EDFA C-band (1528-1563 nm) is split into red (1545-1563 nm) and blue bands (1528-1545 nm), the red band being used for the star over ring channels, and the blue band being used for the mesh-over-ring channels. This kind of configuration is useful when the network requires roughly equal numbers of star-over-ring and mesh-over-ring connections.

One flexible configuration is shown in Figure 5, in which all channels are fully demuxed at the core using demultiplexor 502, with an optical switch 504 placed at the demux output for each channel. The switch 504 can then be used to choose whether a specific channel is a star-over-ring or mesh-over-ring channel. If the former, the switch routes the signal to a transponder 506 for termination, with the groomed wavelength added back onto the ring through an optical multiplexer 508. If the latter, the switch 504 routes the signal to a different multiplexer 510 that combines all the mesh-over-ring channels so they can be attenuated by attenuator 512 then amplified by amplifier 514 up to the reference powers. The star-over-ring and mesh-over-ring channels can then be combined with a suitable coupler 516.

Other variations can exist that are optimized for different architectures. For example, one may want to design a system optimized for (1) many star-over-ring channels and few mesh-over-ring channels, or (2) few star-over-ring channels and many mesh-over-ring channels. For case 1, a suitable channel separation scheme can be devised, such as the one illustrated in Figure 6, which uses fiber bragg gratings (FBGs) 602 and optical circulators 604 to separate the star-over-ring and mesh-over-ring channels. The express channels are reflected by the FBGs and then routed through the core attenuator 612 and amplifier 614. The star-over-ring channels pass through the FBGs, and are terminated after the demultiplexer 606. The star-over-ring channels are added back at multiplexor 608, pass through the FBG 616 and are coupled with the express channels using circulator 620. One advantage of this architecture is that it gives a flexible number of express channels to bypass the core. While Figure 6 shows a system where 4 wavelengths bypass the core, there is no intrinsic limitation-if 5 express channels are needed, 1 more grating can be used. An optical circulator produces about 0.6 dB of loss, and each grating generates about 0.4 dB loss. For the 4 express channel case, the total insertion loss of the express channel splitter/combiner is 2 dB insertion.

This loss scales with the number of channels separated, and at some point the band- splitting architecture of Figure 4 becomes more sensible. The circulator-FBG architecture of Figure 6 can be inverted for case 2 above, in which there are a few star- over-ring channels and many mesh-over-ring channels. This is shown in Figure 7. In this case, the signals transmitted through the FBGs pass through the core attenuator 712 and amplifier 714, and the reflected signals are demultiplexed and terminated. Different designs from those shown can be invented-for example, one could use interleavers to separate even and odd channels.

Up to now, the disclosure has focused on the architecture of a single closed loop amplified ring. In practice, most networks will have at least two rings with working and protection paths. Typically, a working path is the shortest connection between two nodes; in the case of star-over ring, this means the shortest connection between the core and an edge node. However, in the closed ring amplified architecture, it can be preferable to define the working path as the path that does not go through the reference node or the core. In other words, the express channel can be viewed as the protection path. This definition of working path minimizes the number of express channels on the ring when protected services are not needed. In fact, in the mixed star-over-ring/mesh- over-ring architecture, express channels are not required at all for unprotected services.

This simplifies the network and eliminates the core amplifier, which ultimately results in a reduction of cost. In the full mesh-over-ring architecture, there will always be some express channels on the ring, because any channel terminated at the reference node is inherently an express channel. Still, this kind of architecture will minimize the number of express channels on the ring. This is helpful, because express channels are added to the ring with higher power levels than other channels, which is not ideal for amplifiers in the system.

In order for the closed loop amplified system to work, one must ensure that the cavity gain is always less than the cavity loss. In reality, variations and uncertainties in power levels in the system can affect the stability and cause the loop to start oscillating, potentially bringing down the network. The key point to keep in mind is that the power of the reference channel, as it goes around the ring, is used to calculate the ring loss. This is what allows one to set the express channel power, at the add location, to a value higher than the reference power, which in turn is what ultimately prevents the amplifier loop from lasing when the system is properly configured to satisfy the power matching condition at the reference node. As a result, if the span gain or loss differs between reference channel and express channel, or between reference channel and a portion of the ASE spectrum from the amplifier, catastrophic network failure can occur. Some of these power level variations and uncertainties are (1) power differences between the reference channels and the express channels at the reference node; (2) loss variations for those parts of the ring that are not common for the reference channel and its corresponding express channel. For example, in FIG. 4, the power levels of the multiplexer 408 and the core amplifier 406, when exiting the red-blue combiner, should be the same; however, before being combined the signals are on different fibers and inside the combiner one signal is reflected off a filter, and another is transmitted through the same filter. So although the output of the multiplexer 408 and the core amplifier 406 may be the same, the fact that the path losses are different could result in less round trip loss for the express wavelengths, even to the point where the round trip gain minus loss is greater than zero; (3) miscalibrations of power meters which are used to measure the reference channel powers and the added express channel powers at an add-drop site; (4) gain tilt in amplifiers which causes the reference channel gain to be different from the express channel gain, or the ASE gain in an unused portion (no channel there) of the amplifier spectrum; (5) wavelength dependent loss in passive components throughout the system, that results in different ring loss for reference and express channels (or ASE).

Catastrophic failure can occur when the span loss for an express channel is less than the span loss for its reference channel, or when the amplifier gain for an express channel is greater than the gain for its reference channel. A similar situation occurs if the reference channel power originating at the reference node is lower than the express channel power at the reference node. In addition to the above scenario, it is also possible that the ASE in an unused portion of the amplifier gain spectrum could experience higher gain around the ring than the region used by the active channels. It is not uncommon for an amplifier to have a gain tilt of 2 dB across its gain bandwidth. As a result, if the cavity loss is being calibrated with a reference channel in the low gain portion of the spectrum, the high gain portion may lase, or at least generate excessive amounts of ASE. The severity of the problem (as with the problem discussed above) depends on the margin of required cavity loss that is designed into the system. If a cavity with 2 dB gain tilt is designed to operate with 1 dB cavity loss at the LOW gain point, this ring will certainly lase in the high gain portion of the spectrum. On the other hand, a design that generates 5 dB of cavity loss will suppress lasing and excessive ASE at the high gain end also. Chaining multiple amplifiers with gain tilt in this architecture either requires an increase in the marginal cavity loss (loss minus gain around the ring), or requires restricting the reference channels to lie only within the high gain portion of the amplifier.

Increasing the marginal cavity loss is equivalent to increasing the power difference between express channel and reference channel when an express channel is added to the ring. The potential problem with this is that the subsequent amplifiers through which express, reference, and normal channels pass must be able to tolerate the nonuniformity in input power. While this may be tolerable in architectures with relatively few express channels, it could be problematic in architectures with a large number of express channels. One solution would be to build very high output power amplifiers and operate them only in the linear region. Another solution would be to tailor the channel plan so that all reference channels occupy the high gain regions of the amplifier spectrum. High precision control and monitoring of transmitter and amplifier power levels would be required along with this.

The present invention has been described in terms of specific embodiments, which are illustrative of the invention and not to be construed as limiting. Other embodiments are within the scope of the following claims.

What is claimed is: