Modeling Open Optical Line Systems with GNpy

Bill Owens, Chief Network Architect, NYSERNet

Why Open Optical?• Open is better than closed!

• You can understand your own network and use it most effectively.

• Vendors can create new components, software and services that provide incremental improvements without wholesale replacement.

• You don’t really own your network unless you can understand, control and modify it.

• The community can contribute to development and support.

UnderstandVendor’s optical design tool: this network has 96 active channels at 50 GHz and 0 dBm per channel.

Reality: this network has 2 active channels, and because of the design tool they’re each at 0 dBm so their SNR is much worse than it could be.

And the tool tells us our network won’t support a particularly demanding path, but when we plug in the equipment it works fine.

ImproveVendor’s network design: our amps connect to our ROADMs that connect to our line cards. Our management software assumes that it will only ever deal with our components.

Reality: The analog line system is from Vendor A, along with some of the transponders. Some other transponders come from Vendor B because they’re cheaper. Some from C because they’re more flexible, and one path is from Vendor D to E since they make compatible optics.

Dream: Improved amps and upgraded ROADMs from third-party vendors make the line system perform better, and are integrated through open management interfaces.

Voyager PoC path includes three vendors (ADVA, Ciena and Cisco) and direct optical handoff between networks.

Community

If we open the optical technology that we all use, we can help each other.

Gaussian Noise Model

If we can make some simple assumptions about a network, namely:

• The topology consists of fiber spans between amplifiers, without dispersion compensation.

• The transponders are using coherent detection.

Then we can model the “quality” of the network as a simple combination of the noise from the amplifiers (ASE) and from nonlinear effects in the fiber (NLR).

ElementsThe library includes basic network components:

• Fiber

• Amplifiers - EDFA (multiple models) and Raman (under development)

• ROADMs (basic modeling)

• Transceivers

We can improve the quality of the output by improving those models - but to do that we need help from the vendors.

Simple models - low, medium and fixed gain

Detailed model• Basic params - gain, power• Noise tigure, tilt, gain ripple every 50 GHz

EDFA Models

Raman amplifiers• Current Raman calculations take a few seconds per channel.

• Times 96 channels times multiple spans times different scenarios...

• The PSE project is working on a closed-form approximation to enable real-time solving.

Network LayoutNetworks can be described in JSON:

• Fiber spans• Amplifiers• ROADMs• Transceivers

Or in Excel with a JSON translator

ResultsThis is a very simple example with just one fiber span and one EDFA

SNR total = 27.78 dB

ResultsLet’s change that to a 120 km span...

SNR total = 21.86 dB

Where do we go from here?

• The GNpy project is about writing a library, not an application - the demo app is just that, and isn’t intended to be a network planning tool.

• But we want network planning tools, so we’ll have to write them.

• The Raman approximation needs to be written and tested.

• We need equipment vendors to contribute parameters for their equipment.

• We need operators to model their real networks with GNpy!