From Installed Base to AI-Ready Edge: A Practical Migration Framework [6/6]

June 24, 2026

Access networks running on 10G SFF DWDM are running out of capacity. The upgrade to 100G looks simple — but the choice of module matters, the deployment path depends on the link, and sequencing determines whether the program delivers in two quarters or stalls for two years. This paper guides both decisions, from first principles.

Why the Module Choice Is Not Trivial

If your access network runs on 10G SFF DWDM tunable transceivers, you already know the capacity conversation is coming — if it hasn’t started already. Upgrading to 100G looks straightforward from the outside: swap the module. The wrinkle is that several 100G module formats exist, and not all of them work on the C-band DWDM infrastructure that most access backhaul is built on.
Three categories come up regularly in access upgrade discussions:

PAM4 direct detect modules — LR4, ER4, ZR4 — use intensity modulation across O-band or short C-band wavelengths. ZR4 reaches up to 80 km over SMF using four O-band LAN WDM channels with host FEC — a viable option for some point-to-point metro links. The reason it doesn’t fit C-band DWDM access backhaul: it operates in the O-band, so it can’t share C-band DWDM infrastructure, and 80 km is its ceiling.
100G ZR coherent runs on the same C-band fiber, the same ITU-T wavelength grid, the same open line system as the 10G SFF DWDM tunable it replaces. The coherent DSP handles dispersion and phase effects across the full 25–500 km access backhaul range. That’s the technology decision. Everything that follows is about executing it well.

Where 100G ZR Lives in the 5G RAN Architecture

In a disaggregated 5G RAN, the transport network splits into three layers. Fronthaul runs from the Radio Unit at the tower to the Distributed Unit — typically within 25km, carrying eCPRI traffic with tight latency require-ments. Midhaul and backhaul connect the DU to the Centralized Unit at the MSO aggregation hub, across spans of 25 to 500km with more room on latency. That midhaul/backhaul tier is exactly where 100G ZR operates.

Figure 1: : 5G RAN transport layers — 100G ZR QSFP28 operates on the midhaul/backhaul segment between the DU at the cell site and the CU/Hub at the aggregation point. Sources: 3GPP TS 38.401; O-RAN Alliance.

Not every access link is a 100G ZR candidate — fronthaul is a different conversation. And those DU nodes are increasingly AI inference hosts: the backhaul link now carries model updates, split inference results, and CMIS telemetry alongside user traffic. Getting this link right matters well beyond the capacity upgrade.

The Decision Filter: Four Criteria Before Committing

Before any purchase order goes out, it’s worth spending an hour with four questions per candidate link. In most access networks, 60–70% of links sail through all four. The rest are either better served by a different approach or simply not ready yet — and knowing that upfront saves a lot of program pain later.

Figure 2: Operator self-assessment decision filter — four criteria determine whether 100G ZR drop-in, a forklift evaluation, or deferral is the right path for each link. Source: Arycs Technologies.

Distance and dispersion. C-band DWDM access backhaul spans roughly 25 to 500km. Below 25km you’re in fronthaul territory. Above 500km you’re outside the 100G ZR envelope without additional signal processing. The good news: the overwhelming majority of mobile backhaul and aggregation rings sit comfortably within that window.
Traffic growth trajectory. An 18-month forecast is the right horizon. Links already pushing 70% utilization will create service problems before a deferred migration can complete — those are your Phase 1 candidates. Links sitting below 50% with flat growth can wait. The temptation to sequence by geography rather than congestion is real, especially once procurement and field ops get involved. This check is what keeps the program honest.
For spans beyond 120 km, an existing C-band EDFA OLS is required. If one’s there — which it is for most access backhaul at that distance — no new hardware needed; the module adds a wavelength to the grid. No OLS? That’s the prerequisite, not the module.
Host compatibility. Two quick checks: QSFP28 port available, thermal budget in the cabinet adequate. Both take minutes. Surprisingly, this is where Phase 1 pilots most often hit trouble — not the optics, the host. Easy to fix once found; expensive to discover after procurement.

Two Deployment Paths Within the Drop-in Route

Links that pass the decision filter split into two commissioning paths based on span length. Most access networks are roughly 60/40between them. The two paths are genuinely different in what they require — treating them the same is the single most reliable way to generate Phase 1 delays.

Unamplified (up to 120 km). Connect directly to dark fiber or a passive C-band DWDM MUX. No EDFA, no OLS integration. The majority of metro backhaul and short aggregation rings land here. It’s the natural Phase 1 starting point — cleanest commissioning, fastest path to field-validated results.

Figure 3: Unamplified 100G ZR deployment — single span up to 120 km on G.652 SMF. No amplification required.

Amplified (120–500 km). The module joins the existing C-band OLS as a new wavelength client. No new amplifier hardware if the OLS is already there. What’s different from the unamplified path is the commissioning: you need to validate the OSNR budget against the amplifier chain and confirm wavelength assignment on the grid before the link goes live. Skipping this because “the OLS is already working” is a pattern we see regularly — and it’s consistently the reason a pilot that should take days ends up taking weeks.

Figure 4: Amplified 100G ZR deployment — 120–500 km across existing C-band OLS with inline EDFA amplifiers. No new amplifier hardware required.

Three Phases — Commitment Follows Field Evidence

Phase 1 — Pilot Validation
Pick 3–5 of your highest-congestion links, not the most accessible ones. Lightly loaded pilot sites hide the constraints that matter — thermal limits, NMS gaps, OSNR margins tighter than the survey suggested. Deploy in SFF-8636 mode on existing hosts: no upgrade needed yet. Go/no-go into Phase 2 is field results, not data sheets.

Phase 2 — Congestion-Priority Rollout
Expand to every link forecast to breach 70% within 18 months, ordered by congestion severity. The most common Phase 2 story: sequencing quietly reverts to geographic clustering once field ops get involved. Lock in congestion-priority ordering before rollout begins. Legacy 10G modules stay in adjacent slots for per-link rollback throughout. Get coherent visibility into your NMS before deployment density grows.

Phase 3 — CMIS Unlock and AI Readiness
When a host platform gets modernized to CMIS-capable management, firmware-switch the deployed modules from SFF-8636 to CMIS mode — no module swap, no outage. What that unlocks is the full physical layer telemetry dataset: per-channel OSNR, pre-FEC BER margin, dispersion margin, receiver sensitivity headroom — all flowing to the AI management layer in real time. The one thing that catches teams out here is treating the CMIS enablement as a follow-on task after the host upgrade. Plan them together. ‘We’ll sort out CMIS later’ has a strong track record of meaning never.

The Conversation Worth Having First

The technology works and the economics are solid. What separates a smooth migration from a painful one is sequencing discipline at the start, validation rigor at each gate, and a clear NMS integration model.

The failure patterns are consistent enough that they’re almost predictable: Phase 1 pilot sites chosen for access not congestion; Phase 2 sequencing that drifts geographic by week three; Phase 3 CMIS planning pushed to “after go-live” until it falls off the program entirely. None of these show up in a project plan. All of them are avoidable. The difference is usually a conversation that should have happened before the first purchase order was raised — and that’s a conversation worth having.

#MigrationFramework    #CoherentAccess    #100GZR    #AI-RAN

About Arycs Technologies

Arycs delivers power-efficient, coherent-class optical connectivity based on silicon photonics, coherent DSP, and advanced optical architectures. Our solutions provide industry-leading bandwidth per watt, deterministic performance, and flexible network evolution for AI, cloud, telecom, and edge infrastructure. Designed for real-world deployment, Arycs Technologies enables networks to scale with growing AI demand without disruptive redesign or hardware replacement.

arycs-tech.com  |  LinkedIn: Arycs Technologies

From Installed Base to AI-Ready Edge: A Practical Migration Framework [6/6]