1. Introduction

A wavelength service ends at a client port. A spectrum service ends at a frequency slot, and everything downstream of that slot is someone else's equipment operating inside your fiber. That single change of boundary is what makes optical spectrum as a service (OSaaS) a different engineering problem from every other transport product an operator sells. The line system still amplifies, still equalises, still carries the tenant's light through the same erbium and the same silica as its own channels, but it no longer knows how many carriers are in the slot, what their modulation format is, or what happens when the tenant's network operations centre decides to drop six of eight channels at two in the morning.

The commercial pull is obvious enough. A regional operator with a lit C-band route and forty percent spectral occupancy has an asset it cannot monetise as wavelengths, because the buyer wants control of its own transceiver refresh cycle and does not want to renegotiate every time it moves from 400G to 800G. Selling 400 GHz of contiguous spectrum on a fifteen-year term converts idle bandwidth into revenue while leaving the transponder economics with the party that cares about them. Research and education networks got there first, for the same reason they got to alien wavelengths first: their traffic growth outruns their capital budgets, and their peers are trusted counterparties.

The engineering pushback is equally concrete, and it has a number attached. In a survey of twenty-five network operators reported in doctoral work at Tallinn University of Technology, sixty percent named power and power spectral density management inside the leased window as their primary concern, and others specifically named the risk of a customer injecting intensity-modulated on-off keyed (OOK) signals into a line system engineered for coherent channels. Both concerns describe the same failure: a tenant does something entirely reasonable inside its own slot, and a channel three hundred gigahertz away, belonging to somebody else, drops below its forward error correction (FEC) threshold.

That coupling is not a bug in anyone's implementation. Erbium-doped fiber amplifiers (EDFAs) share gain across the whole band. The Kerr nonlinearity in the transmission fiber couples intensity in one channel to phase in its neighbours. Stimulated Raman scattering tilts power from short wavelengths to long ones across the entire occupied spectrum. None of these mechanisms respects the frequency boundaries drawn on a commercial contract, and all of them are the reason a spectrum service needs a demarcation architecture rather than a spreadsheet of allocated slots.

This article works through that architecture in four parts. First, what a handoff specification has to contain to be enforceable — the parameters, their reference points, and why power spectral density rather than total power is the term that actually binds. Second, where the demarcation physically lives and what the three practical interconnect options give up. Third, how per-tenant monitoring works when the operator is deliberately blind to the tenant's internal channel plan, including the optical channel monitor (OCM) technique that recovers 6.25 GHz resolution inside an opaque window without adding a filtering penalty. Fourth, the policing mechanisms — detection, attribution, and action — along with published measurements of how accurate and how fast they are.

The treatment assumes working familiarity with coherent DWDM engineering. Readers who want the underlying noise accounting first should start with the OSNR fundamentals primer, and those who want the architectural context for open line systems should read how open line systems carry multi-vendor coherent wavelengths before continuing here.

Takeaway: A wavelength service hands over a bit stream and keeps the physics. A spectrum service hands over the physics and keeps the liability. Every mechanism in this article exists to close that gap — to give the operator enough observability and enough authority at the boundary that a tenant's freedom inside its slot stops at the edge of its neighbour's.

2. Spectrum Service Definition and Scope

2.1 Service Tiers and Responsibility Boundaries

Transport products differ in one variable: how far down the stack the demarcation sits. Dark fiber puts it at the fiber connector; the buyer owns the amplifiers, the ROADMs, the channel plan, and every decibel of the loss budget. A managed wavelength puts it at a client-side Ethernet or OTN port; the buyer owns nothing optical. Spectrum sits between them, and a managed optical fiber network (MOFN) sits between spectrum and dark fiber, with the seller running the line system as a service on the buyer's behalf.

The distinction that matters for engineering is not who pays for what — it is which party's decisions can change the physical layer. On dark fiber, the buyer's decisions change only the buyer's fiber. On a wavelength, the buyer has no physical-layer decisions to make. On spectrum, the buyer's decisions change a shared fiber that other parties are also using, which is the only one of the four cases where a purely commercial boundary sits inside a coupled physical medium.

Responsibility split across four optical transport service tiers A four-column comparison showing which party owns the fiber, amplifiers and ROADMs, the channel plan, the transceiver and the launch power for dark fiber, managed optical fiber network, optical spectrum as a service and managed wavelength services. Spectrum-as-a-service is the only tier where the tenant controls launch power inside a fiber that is shared with other tenants. Responsibility allocation and demarcation position by service tier Blue = provider responsibility. Amber = tenant responsibility. The shaded band marks decisions that couple into a shared fiber. Layer of responsibility Dark fiber MOFN Spectrum service Managed wavelength Fiber and duct Provider Provider Provider Provider Amplifiers and ROADMs Tenant Provider Provider Provider Channel plan in the slot Tenant Provider Tenant Provider Transceiver Tenant Tenant Tenant Provider Launch power into a shared fiber Not shared Provider Tenant policing needed here Provider Demarcation point Fiber connector Coherent interconnect Frequency slot Client port Only the spectrum tier places a tenant-controlled launch power inside a fiber whose amplifiers and nonlinearity are shared with other tenants. That is the entire source of the problem.
Figure 1: Responsibility split across four transport service tiers. The shaded column is the only one where a commercial boundary is drawn inside a physically coupled medium.

2.2 Frequency Slot Definition per ITU-T G.694.1

A spectrum service delivers a frequency slot as defined by ITU-T G.694.1 — a nominal central frequency and a slot width, anchored to 193.1 THz. Under the flexible grid, the nominal central frequency granularity is 6.25 GHz and the slot width granularity is 12.5 GHz, so a slot is described as 193.1 THz plus n times 6.25 GHz, with a width of 12.5 times m gigahertz (standard-specified, ITU-T G.694.1). A 400 GHz allocation is m = 32. The tenant may place any carriers it likes inside that slot, at any baud rate and any modulation format, provided the composite signal stays inside the slot edges and inside the agreed power envelope. Everything about the grid mechanics is covered in more depth in the G.694.1 channel-grid walkthrough.

What the tenant does not buy is a guarantee of quality of transmission that is independent of its own behaviour. This is the point most commercial teams get wrong on the first contract. The generalised signal-to-noise ratio (GSNR) delivered at the far end is a function of the launch power the tenant chose, the channel count the tenant configured, and the modulation formats the tenant selected — none of which the operator controls. A spectrum service-level agreement that promises a delivered OSNR without constraining the tenant's launch conditions is unenforceable in both directions.

Engineering note

Write the spectrum SLA as a conditional. The operator warrants a minimum delivered GSNR provided the tenant operates within a stated power spectral density envelope, a stated per-carrier bandwidth floor, and a stated modulation class. Outside that envelope the warranty lapses and the policing rules apply. Anything else asks the operator to guarantee an outcome it cannot observe and cannot influence.

2.3 Technology Enablers for Commercial Adoption

Three things changed. Coherent transceivers became commodity pluggables, so a tenant can light a slot with a router line card instead of a transport shelf. Flexible-grid wavelength selective switches (WSS) became ubiquitous, so an operator can carve a contiguous 300 or 400 GHz block without wasting a fixed-grid channel plan. And streaming telemetry over gNMI replaced polling, so per-second visibility into a probe channel's performance became a normal operational capability rather than a lab exercise — the pattern described in the network-as-code approach to optical automation.

What has not changed is the trust model. Spectrum services today are still overwhelmingly sold to counterparties the operator already trusts — peer carriers, national research and education networks, large enterprises with their own transport teams. The technical work described in the rest of this article is what would be needed to widen that circle, and the fact that it is still an active research subject rather than a settled product feature tells you where the model actually stands.

Takeaway: Spectrum is the only transport tier where a commercial boundary is drawn inside a coupled physical medium. Every other tier either isolates the buyer on its own fiber or keeps the physical layer entirely with the seller. That is why spectrum needs a demarcation function and the others do not.

3. From Alien Wavelength to Alien Spectrum

3.1 Alien Wavelength Operation and Control

An alien wavelength is a single channel from a third-party transponder carried across another vendor's line system. The technique is old — European research networks were running production alien wavelengths years before the phrase "open line system" was in vendor marketing — and it works because the line system retains complete authority over the channel. The multiplexer add port constrains the channel to its assigned centre frequency and bandwidth through hard optical filtering, and the line system monitors and attenuates the channel's power as part of its normal per-channel levelling loop. Ribbon describes exactly this in its open line system documentation: the third-party transponder is registered as an unmanaged element and the channel is defined as a virtual transceiver on the port, with declared transmit power, chromatic dispersion tolerance and polarisation-mode dispersion tolerance, after which it receives the same optical performance monitoring and restoration treatment as a native channel (vendor description, Ribbon).

The critical property is that the alien channel is one channel. The line system knows its bandwidth, knows its power, and can adjust that power independently of everything else. The per-channel variable optical attenuator (VOA) inside the WSS gives the operator a control handle that acts on exactly the misbehaving entity. That precise correspondence between the unit of monitoring and the unit of control is what makes alien wavelengths safe, and it is exactly what alien spectrum breaks.

3.2 Loss of Per-Carrier Control in Block Allocation

Extend the alien wavelength concept to a block of spectrum and the line system's control handle becomes an attenuator across the whole window. It can hold the tenant's total window power at a target. It cannot see, and therefore cannot separately act on, the carriers inside. Ribbon's own description of the failure is the clearest short statement of the problem in public vendor literature: a customer leases 200 GHz, declares four channels spaced 50 GHz apart at 0 dBm each, the operator sets the line system to hold 6 dBm across that 200 GHz band, and then the customer launches a single wavelength. That one carrier gets amplified to the full 6 dBm the operator budgeted for four, and the resulting power density drives cross-phase modulation into other parts of the spectrum (vendor description, Ribbon).

Note what did not go wrong there. The customer did not exceed the total power limit — it stayed exactly at it. The operator's control loop worked exactly as designed — it held the window at the agreed target. The contract was honoured on both sides. And the neighbours degraded anyway, because the term both parties agreed on was the wrong term.

Practical Example — four-carrier to single-carrier power concentration

Four carriers at 0 dBm each in a 200 GHz window give a composite of 10·log₁₀(4) = 6.02 dBm, and an average power spectral density of 6.02 − 10·log₁₀(200) = −17.0 dBm/GHz. Drop three carriers and hold the window power at 6.02 dBm, and the surviving carrier now runs at 6.02 dBm rather than 0 dBm — a 6 dB increase in per-carrier power. If that carrier occupies 50 GHz, its in-band power spectral density is 6.02 − 10·log₁₀(50) = −11.0 dBm/GHz, a 6 dB rise over the design value. Under Gaussian-Noise-model scaling, nonlinear interference power grows as the cube of channel power, so a 6 dB rise in launch power raises nonlinear interference by roughly 18 dB while the useful signal rises only 6 dB — a net 12 dB degradation in the nonlinear-limited signal-to-noise ratio contributed to every channel within the interaction range. The tenant did nothing the contract forbade.

3.3 Point-to-Point and Switched Spectrum Services

Commercially, spectrum services come in two shapes, and the difference is whether the slot survives a ROADM. A point-to-point spectral pipe hands the tenant a fixed slot between two endpoints; the line system carries it end to end without switching it. A switched or sliced spectrum product lets the tenant's block traverse ROADM nodes and be routed as a virtual optical network — Ribbon calls the two varieties spectral pipes and spectral slicing (vendor description, Ribbon). In submarine systems the equivalent product partitions a fiber pair's spectrum among end users so each sees what Ciena describes as a virtual fiber pair, an approach practical mainly on newer uncompensated cables with wide repeater bandwidth (vendor description, Ciena).

The submarine case is worth a sentence of caution because it is often cited as proof that terrestrial spectrum sharing is a solved problem. It is not the same problem. Submarine spectrum sharing operates on the principle of maintaining a fixed total power across the whole fiber, which the repeater chain is designed around and which no single tenant can perturb without the constant-total-power control absorbing it. Terrestrial open line systems with ROADMs, add/drop at intermediate nodes and per-channel levelling do not have that structure, and the submarine control philosophy does not transfer cleanly.

Takeaway: Alien wavelengths are safe because the unit of monitoring and the unit of control are the same object — one channel, one VOA. Alien spectrum breaks that correspondence: the operator monitors and controls a block while the tenant operates carriers. Every technique in the rest of this article is an attempt to restore the correspondence without requiring the tenant to disclose its channel plan.

4. Handoff Specification and Reference Points

4.1 Reference Points per ITU-T G.698.2 and G.Sup39

ITU-T already has a vocabulary for handing an optical signal between parties, and a spectrum service should use it rather than invent one. G.698.2 specifies amplified multichannel DWDM applications with single-channel optical interfaces — the black-link model, where a transmitter from one vendor and a receiver from another are specified at reference points Ss and Rs, and everything between them is treated as an opaque link with defined path penalties (standard-specified, ITU-T G.698.2). G.Sup39 sets out the compatibility taxonomy that sits behind it: transverse compatibility means the two ends of an optical section may be terminated by equipment from different manufacturers, which requires a complete parameter set at both interface points; longitudinal compatibility means both ends come from the same manufacturer and a much smaller parameter set suffices (standard-specified, ITU-T G-series Supplement 39).

A spectrum service is a multi-span, partially transverse-compatible arrangement in that taxonomy, and G.Sup39 is candid that the multi-span transversely compatible cases were considered and deliberately left unspecified when G.691 was first published. It offers instead the notion of joint engineering: the process by which operators agree a set of interface characteristics when the available Recommendations are not sufficient to guarantee the required performance, with the explicit observation that every such situation is different and standardising the parameter values would be meaningless (standard-specified, ITU-T G-series Supplement 39). That is a precise description of what a spectrum handoff sheet is. The standard tells you the process, and the numbers are yours to negotiate.

Two parameters from G.698.2 carry directly into the spectrum contract and should be named in it. Receiver OSNR tolerance is the minimum optical signal-to-noise ratio required to achieve the target bit error ratio at a receiver reference point at a given power level in an OSNR-limited amplified system, and it is explicitly a design parameter rather than a measured one. Optical path OSNR penalty is its companion for the link (standard-specified, ITU-T G.698.2, as summarised in G.Sup39). Both give the two parties a shared way to say what "good enough at the far end" means.

Reference points and parameter sets at a spectrum service handoff A left-to-right diagram showing the tenant's transmit domain, the ingress demarcation point, the provider's open line system with amplifiers and ROADMs, the egress demarcation point and the tenant's receive domain. Parameters enforced at ingress are listed below the ingress point, parameters warranted at egress are listed below the egress point. Handoff reference points: ingress enforcement and egress warranties Tenant transmit Own transceivers Own mux or edge WSS Own channel plan Ingress demarcation Reference point Ss Filter + monitor + VOA Provider open line system EDFA chain, spans, ROADM degrees Provider channels sharing the same gain Other tenants sharing the same fiber Egress demarcation Reference point Rs Drop + monitor Tenant Rx Enforced at ingress — the tenant must comply Slot: nominal central frequency and slot width per G.694.1 Maximum composite power across the slot, in dBm Maximum power spectral density, in dBm/GHz Minimum power spectral density, in dBm/GHz Permitted signal class: coherent only, no intensity modulation Minimum per-carrier occupied bandwidth and guard band at slot edge Warranted at egress — conditional on compliance Minimum delivered GSNR across the slot Maximum accumulated chromatic dispersion, in ps/nm Maximum differential group delay, in ps Delivered power window at the drop port, in dBm Spectral flatness across the slot, in dB Maximum filter narrowing from cascaded WSS passbands Enforcement asymmetry between ingress and egress terms Ingress terms are observable and enforceable at the demarcation point. Egress terms are warranties that lapse if ingress terms are breached. Never warrant an egress number unconditionally. Commonly omitted parameter: minimum PSD Minimum power spectral density. Without a floor, a tenant that under-fills its slot lets the provider's gain-flattening loop redistributes erbium gain toward the channels that remain, tilting the band.
Figure 2: Reference points and the parameter split at a spectrum handoff. Ingress terms are enforceable because they are observable at the demarcation point; egress terms are warranties conditional on ingress compliance.

4.2 Handoff Parameter Sheet

Published OSaaS work converges on a compact handoff sheet. The Open Ireland testbed studies define, for each tenant, a start frequency, an end frequency, an optical bandwidth, a maximum power and a maximum power spectral density — with the power and PSD limits derived from the equipment constraints of the line system and the service-level agreement with the customer, and a margin added to absorb power fluctuation and loss variation (measured study configuration, Raj, Kilper and Ruffini). In their three-tenant experiment each tenant received a 400 GHz window with a maximum power of 10 dBm and a maximum PSD of −15 dBm/GHz, with a ±1 dBm margin, derived from the assumption of eight 200 Gbit/s 16QAM channels at 0 dBm launch and 50 GHz width inside the window.

Premium Article — Free 21% Preview

Read the Full Analysis with Premium

The remaining 79% of this article — the design numbers, trade-offs and field guidance — is part of MapYourTech Premium, along with the full premium library, courses and professional tools.

Instant access · Cancel anytime · 48-hour trial available
Sanjay Yadav

Optical Communications & Network Automation Expert | Author of 3 Books for Optical Engineers | Founder, MapYourTech

Optical networking engineer with nearly two decades of experience across DWDM, OTN, coherent optics, submarine systems, and cloud infrastructure. Founder of MapYourTech.

Follow on LinkedIn

Related Articles on MapYourTech