Can GPON Use a 1:128 Split Ratio? Standards & Limits

The short answer

GPON's TC-layer design considers a 1:128 split, and some OLTs support 128 registered ONUs per port; neither fact automatically approves a physical 1:128 ODN. Treat 1:128 as an engineered exception, not a default. Before release, the design should pass all five gates below:

1. Equipment gate: the exact OLT line card, optical module, software release and ONU combination must support 128 registered ONUs per port.
2. Optical gate: the worst optical path-not the average path-must remain within the applicable upstream and downstream loss limits with an explicit engineering margin.
3. Component gate: the calculation must use the supplier's guaranteed maximum insertion loss and worst-port uniformity, not a theoretical 21 dB split or a typical catalogue value.
4. Capacity gate: the expected connected, busy-hour and simultaneously active subscriber counts must fit the service tiers and DBA policy.
5. Operations gate: the larger fault domain, test access, port records, spares and migration plan must be acceptable to the operator.

If one gate fails, an OLT data sheet that says "128 clients" does not rescue the design. Glory Optical's FTTH network design guide maps these checks into the ODN architecture, loss-budget sheet and BOM before the ratio is frozen.

1. What the GPON standards actually say about 1:128

The most frequently repeated statement about 1:128 is also the one most often stripped of its context.

ITU-T G.984.1 (03/2008), clause 12 says that split ratios up to 1:64 were realistic for the physical layer with the technology considered at publication. The very next sentence says that, anticipating continued optical-module evolution, the transmission convergence, or TC, layer must consider split ratios up to 1:128. "Must consider" defines the TC-layer design scope; it is not a statement that every GPON optical interface supports a 1:128 ODN.

That distinction explains why an OLT manufacturer may advertise 128 clients per PON port while an operator still rejects a 1:128 ODN. Ubiquiti's model-specific UF-OLT guide states support for up to 128 clients per GPON port. An older ZTE platform article, published in 2010, states a 1:128 maximum for the ZXA10 C300. These examples show that equipment implementations have claimed this scale; they are not current procurement approval for another model, release or optical module. Verify the active vendor documentation before design release.

GPON line rates do not increase with the split ratio

The widely deployed asymmetric GPON profile operates at 2,488.32 Mbit/s downstream and 1,244.16 Mbit/s upstream. The current ITU-T G.984.2 PMD specification defines these nominal rates and physical-layer parameters. A 1:128 splitter creates more endpoints; it does not create more PON capacity.

B+, C+ and D are loss classes, not marketing adjectives

GPON optical class	Minimum path loss	Maximum path loss	Practical meaning for 1:128
B+	13 dB	28 dB	Usually too small once a real 1×128 splitter and design margin are included
C+	17 dB	32 dB	Possible for short, clean paths with controlled components
D	20 dB	35 dB	More optical room, but equipment compatibility and economics still require review

The document structure matters. G.984.2 Annex A, which forms an integral part of the Recommendation, identifies B+ as industry best practice for the 2.488/1.244 Gbit/s system and gives a 13–28 dB loss range in Table A.2. Appendix V explicitly does not form an integral part of the Recommendation; it presents single-sided reach-extended C+ and D industry practice, with 17–32 dB and 20–35 dB ranges in Tables V.3 and V.4.

These are published reference budgets, not proof that every quoted OLT/ONU pair implements the class. Do not approve a link merely because a quotation says "C+," "C++" or "high power." The RFQ should state the exact minimum launch power, receiver sensitivity, overload limit, supported loss class, wavelength and FEC assumptions for both directions.

Glory's 1:32 versus 1:64 PLC splitter guide is the useful next step when the standard permits several ratios but the project needs a lower-risk baseline.

2. How much optical loss does a 1:128 splitter really add?

The ideal equal-split loss is:

For 128 outputs:

That 21.07 dB is a physics floor, not an orderable product specification. It excludes waveguide excess loss, fibre coupling, packaging, connectors, wavelength dependence, polarization dependence and channel-to-channel variation.

Published 1×128 PLC splitter data illustrates the gap. A Precision Optical Technologies data sheet specifies less than 23.8 dB insertion loss and less than 2.5 dB uniformity for 1×128. An FS ABS splitter data sheet specifies no more than 24 dB insertion loss and no more than 3.0 dB uniformity. A defensible preliminary budget should therefore reserve about 23.8–24.0 dB for the splitter alone, then replace that planning value with the selected BOM's guaranteed maximum.

Why worst-port loss matters more at 1:128

Uniformity is the difference between the lowest-loss and highest-loss outputs. At 1:128, a 2.5–3.0 dB spread is large enough to make one group of ONUs look healthy while the weakest ports sit near the failure boundary. A sample report that shows only an average hides this risk.

For a project-specific PLC splitter, specify at least:

• maximum insertion loss for every port at the required wavelengths;
• uniformity, PDL, WDL, return loss and directivity limits;
• connector type, polish and whether connector loss is included;
• operating-temperature range and test conditions;
• serial-number or batch traceability;
• per-port test data, not only a pass/fail statement.

Single-stage versus cascaded 1:128

All ideal architectures with a total ratio of 1:128 have the same 21.07 dB mathematical split. Real device combinations do not have the same excess loss or operational consequences.

Architecture	Published device-limit example	Connectorized interstage allowance	Planning observation
One 1×128	23.8–24.0 dB	None	Fewest passive stages, but a large module and 128-output fault domain
1×2 + 1×64	3.8 + 20.5 = 24.3 dB	About 0.4 dB	Easy two-branch layout, but little optical advantage
1×4 + 1×32	7.2 + 16.5 = 23.7 dB	About 0.4 dB	Supports phased distribution and can be close to one 1×128 in device loss
1×8 + 1×16	10.3 + 13.5 = 23.8 dB	About 0.4 dB	More local granularity; records and access discipline become critical

The device values are illustrative sums from published maximums, not a substitute for the selected suppliers' specifications. A fusion-spliced interstage may contribute less loss than a connector pair, while a field connector can contribute much more when contaminated or damaged.

3. Worked 1:128 GPON optical-budget scenarios

A useful budget must be reproducible. The following Glory engineering model is an illustrative design calculation, not a claim about a particular live network and not a universal acceptance limit.

Assumptions:

• 1×128 splitter maximum planning loss: 24.0 dB
• Four connector pairs at 0.4 dB each: 1.6 dB
• Six fusion splices at 0.1 dB each: 0.6 dB
• Fibre attenuation: 0.35 dB/km
• Engineering allowance for ageing, repair and unmodelled variation: 3.0 dB

OLT-to-ONU distance	Calculated ODN loss	B+ margin (28 dB)	C+ margin (32 dB)	D margin (35 dB)
2 km	29.90 dB	−1.90 dB	+2.10 dB	+5.10 dB
5 km	30.95 dB	−2.95 dB	+1.05 dB	+4.05 dB
10 km	32.70 dB	−4.70 dB	−0.70 dB	+2.30 dB
15 km	34.45 dB	−6.45 dB	−2.45 dB	+0.55 dB

 

This table shows why "GPON supports 1:128" is not an optical-budget answer. Under these conservative assumptions, B+ fails even at 2 km. C+ is usable only on the shorter examples, and the 5 km case has just 1.05 dB remaining. Class D provides more room but still reaches only 0.55 dB at 15 km-too little for many operators' design policy.

The zero-kilometre test: reject impossible designs before route engineering

Remove fibre distance from the model. The fixed loss is already:

The assumed 1:128 BOM cannot meet a 28 dB B+ budget even at zero kilometres. This check prevents teams from optimizing a cable route when the splitter, interfaces and margin have already consumed the entire class.

The same model can estimate a theoretical distance boundary:

• B+: no positive distance under these assumptions;
• C+: approximately 8.0 km;
• D: approximately 16.6 km.

These are calculation boundaries, not deployment recommendations. Field acceptance still needs measured loss in both directions or at the specified wavelengths, connector inspection, OTDR records where appropriate, and OLT/ONU power verification. Glory's ODN deployment guide provides the broader component and field-work context for that acceptance plan.

Why a calculated pass can still fail in service

• the splitter's highest-loss port rather than its average port;
• connector loss excluded from the splitter data sheet;
• reflectance or a dirty end face that does not appear as simple attenuation;
• the lower-power ONU at the longest branch;
• temperature and ageing assumptions already included-or not included-in component limits;
• a future repair closure and extra splice;
• OLT optical-power reporting uncertainty;
• differences between registration sensitivity and stable error-free operation.

4. Bandwidth limit: what 128 active ONUs actually share

Optical margin answers whether bits can cross the ODN. It does not answer whether the PON can carry the offered traffic.

Using the common 2,488.32/1,244.16 Mbit/s GPON rates and simple equal division, the raw share is:

Simultaneously active ONUs	Raw downstream per ONU	Raw upstream per ONU
32	77.76 Mbit/s	38.88 Mbit/s
64	38.88 Mbit/s	19.44 Mbit/s
128	19.44 Mbit/s	9.72 Mbit/s

 

These are not customer speed guarantees. They are arithmetic ceilings before GPON framing, management traffic, protocol overhead, service policy and burst scheduling. Real GPON relies on statistical multiplexing and dynamic bandwidth allocation, so most users are not continuously active at peak rate. That makes 1:128 commercially possible in some networks-but also makes take-rate and busy-hour assumptions part of the engineering design.

Homes passed, ONUs connected and ONUs active are different numbers

A 1:128 physical tree may pass 128 premises while only 50 subscribe. Of those, perhaps 20 are meaningfully active in a given interval. Capacity planning should record all three values, plus service-tier mix and upstream-heavy applications. Treating 128 outputs as 128 simultaneous gigabit users is unrealistic; assuming peak concurrency will always remain low is equally unsafe.

Port-capacity check

1. Forecast take rate at years 1, 3 and 5.
2. Busy-hour downstream and upstream utilization.
3. Peak-to-average ratio by service tier.
4. Committed or assured bandwidth obligations.
5. A split or augment trigger before congestion affects customers.

For networks selling many gigabit or upstream-intensive plans, the better answer is often more PON ports or a move to XGS-PON, not deeper GPON sharing. A Fiber Connect 2025 presentation deck hosted by the Broadband Forum includes a Nokia "Fiber Reality Check 2025" slide, citing an April 2025 Omdia forecast, that describes XGS-PON as aggressively replacing GPON in new deployments and remaining dominant for the next several years. This is conference presentation material, not a Broadband Forum standards report.

The passive BOM should therefore be checked for coexistence and migration, not optimized only for today's OLT-port count. Glory's FTTH network design guide links subscriber forecasts to splitter and fibre allocation.

5. Operational limits that do not appear in the loss budget

A 1:128 PON concentrates more customers behind one OLT port, feeder path and first-stage component. A failure near the root can therefore affect up to 128 services. The financial saving from fewer line-card ports must be compared with the cost and impact of that larger fault domain.

Records become part of optical performance

With cascaded splitting, a port record should identify:

• OLT chassis, slot, PON port and optical-module class;
• first- and second-stage splitter serial number and port;
• enclosure, tray, adapter and splice position;
• route length and baseline event map;
• ONU serial number, branch loss and commissioning receive power;
• reserved, connected and unavailable outputs.

Without this hierarchy, a technician can measure low power but still not know which upstream component or branch to inspect. Glory's guide to MST, ODF and FDB selection helps align the physical distribution point with the record and maintenance model.

Registration is not the same as stability

An ONU may register at a low optical level during a quiet test, then experience intermittent errors or deregistration after temperature changes, connector disturbance or additional repairs. A practical acceptance plan should compare the measured worst ONU with the operator's warning and alarm thresholds, not only the receiver's absolute sensitivity.

Reliability claims need evidence, not a logo

For passive components, Telcordia GR-1209-CORE provides generic requirements for passive optical components, while GR-1221-CORE addresses long-term reliability assurance. These are valuable qualification references for splitter procurement, especially in outside-plant environments.

However, "Telcordia compliant" should not be accepted as a blanket sentence. Ask which product family, package, issue, sample size, test sequence and report number the claim covers. Then check whether the supplied 1×128 configuration is within that qualification scope. A PLC splitter test report should connect the qualification baseline to per-port production data for the actual shipment.

6. Where a 1:128 GPON split can make sense

1:128 is most defensible when density is high, reach is short, component control is strong and the commercial model values OLT-port consolidation.

Short-reach MDU or campus networks

A large apartment complex, hotel, campus or passive optical LAN can keep fibre distance low and distribution points controlled. A C+ or D design may retain adequate margin if connectors and splitter stages are minimized. In these environments, a fiber distribution box selected for the actual splitter and splice layout is more important than a nominal port count on the enclosure label.

Dense areas with measurable, low initial take rate

A tree may pass many premises while only a fraction subscribe. Cascaded 1:4 × 1:32 or 1:8 × 1:16 distribution can support staged build if the operator installs second-stage capacity as take rate grows. This works only when idle ports, splitter paths and activation records are controlled.

Cost-sensitive networks with an explicit augmentation plan

An operator may accept higher sharing to reduce initial OLT and feeder-fibre cost. The plan should state the utilization, margin or customer-count threshold that triggers a second PON port, a re-split or technology migration. "We will upgrade later" is not a plan unless spare fibres, cabinet space, wavelengths, hardware and outage procedures are reserved now.

7. Where 1:128 is usually the wrong choice

Long rural routes

High splitter loss and long fibre attenuation consume the same finite budget. A rural route also tends to accumulate closures, repairs and environmental exposure. Reducing the split to 1:32 or 1:64 usually buys more useful resilience than saving one OLT port.

High take rate with many premium service tiers

If a port is expected to carry many 500 Mbit/s or gigabit subscribers, the capacity problem may arrive before the optical problem. High upstream demand from cloud backup, video creation, surveillance and business users makes the 1.244 Gbit/s upstream especially important.

Brownfield ODNs with incomplete records

Unknown connector counts, undocumented cascades, mixed polish types and aged closures make a 1:128 calculation look more precise than its source data. Audit and test the ODN first. Glory's high-loss diagnostic guide can serve as the field checklist before adding another splitter stage.

Strict-SLA business or public-service networks

A larger shared fault domain, tighter power margin and higher oversubscription are rarely good trade-offs for services with guaranteed bandwidth or restoration obligations. Use a lower split, protected architecture or dedicated access where the SLA warrants it.

New builds optimized for yesterday's electronics

A 30-year passive plant should not be made inflexible to save a GPON port that may be replaced much sooner. If XGS-PON, 25GS-PON or multi-generation coexistence is on the roadmap, evaluate the future wavelength plan and loss class now. Glory's 25GS-PON optical-budget explainer is a useful migration-side comparison; it is not proof that every GPON 1:128 ODN will support a future PON generation unchanged.

8. 1:32 vs 1:64 vs 1:128 GPON decision matrix

Decision factor	1:32	1:64	1:128
Ideal splitter loss	15.05 dB	18.06 dB	21.07 dB
Typical maximum PLC device range used in planning	About 16.5–17 dB	About 20.5–21 dB	About 23.8–24.0 dB
B+ feasibility	Usually comfortable with a sound ODN	Possible but margin-sensitive	Generally unsuitable with a normal engineering allowance
C+ feasibility	Strong margin in many layouts	Practical in many controlled layouts	Possible mainly on short, clean layouts
Raw bandwidth sharing	Lowest	Medium	Highest
Fault domain per PON port	Up to 32 ONUs	Up to 64 ONUs	Up to 128 ONUs
Record and troubleshooting complexity	Low	Medium	High, especially when cascaded
Best fit	Longer or high-reliability access	Dense mainstream FTTH	Short, dense, carefully controlled exception

 

The extra ideal loss is about 3.01 dB each time the split doubles. In the field, the decision is more than a 3 dB calculation: uniformity, extra interfaces, subscriber concurrency and recovery impact also grow. For a deeper comparison of the common baselines, use Glory's 1:32 versus 1:64 selection guide, then evaluate 1:128 only if neither baseline meets the business objective.

9. What to put in a 1:128 RFQ and acceptance plan

A successful high-split project starts with purchasing language that engineering can test. "1×128 GPON splitter, Telcordia compliant" is not enough.

Glory's guide to reading a PLC splitter test report can be attached to the inspection plan. Once limits are fixed, the Glory RFQ form can carry the splitter configuration, enclosure BOM, test-report requirement and drawing revision as one project package.

Turn an approved 1:128 architecture into a controlled passive BOM

Send the topology, route lengths, optical class, interface details, enclosure plan and test-document requirements. Glory can align the PLC splitter, passive interfaces, enclosure layout and inspection package with the design inputs you approve.

Request a project quotation Review PLC splitter options

10. Frequently asked questions about 1:128 GPON

Q: Does ITU-T allow a 1:128 GPON split ratio?

A: ITU-T G.984.1 requires the TC layer to consider split ratios up to 1:128, while its physical-layer discussion identifies 1:64 as realistic for the technology considered at publication. That puts 1:128 within the TC-layer design scope; it does not give blanket physical-layer approval. The exact OLT/ONU implementation and worst-case optical budget still have to prove the design.

Q: Can a B+ GPON OLT support 1:128?

A: It may register 128 ONUs in software, but a normal 28 dB B+ optical budget is usually too small for a real 1×128 splitter plus connectors, splices, fibre and design margin. In the worked model, fixed loss reaches 29.2 dB before fibre distance.

Q: Is C+ enough for a 1:128 GPON network?

A: Sometimes. With a 32 dB maximum path loss, C+ can support short, clean, tightly controlled 1:128 paths. In the illustrative model, it retains 2.10 dB at 2 km and 1.05 dB at 5 km, but fails at 10 km. Use the actual BOM and operator margin policy.

Q: What is the insertion loss of a 1×128 PLC splitter?

A: The theoretical equal-split loss is 21.07 dB. Public commercial data sheets specify maximum insertion loss around 23.8–24.0 dB for some 1×128 products. The selected supplier's guaranteed worst-port value is the number that belongs in the final budget.

Q: Is one 1×128 splitter better than cascaded splitters?

A: It has fewer passive stages and interstage interfaces, which can simplify loss and records. Cascades can improve phased deployment and local distribution. Neither is universally better; compare worst-case device loss, connector/splice events, enclosure access, testability and fault isolation.

Q: How much bandwidth does each user get on 1:128 GPON?

A: If all 128 ONUs were equally active, the raw arithmetic share would be 19.44 Mbit/s downstream and 9.72 Mbit/s upstream before overhead. Actual service depends on DBA, traffic concurrency and operator policy, so this number is not a guaranteed user speed.

Q: Can an ONU work at −28 dBm?

A: Only if the exact ONU receiver specification and operating conditions permit it, and "works" should not mean merely registering once. Operators normally reserve margin above absolute sensitivity for measurement uncertainty, temperature, ageing, contamination and repair. Check both downstream ONU power and upstream burst reception at the OLT.

Q: What documents should a 1×128 splitter supplier provide?

A: Request the controlled data sheet, drawing, qualification evidence, material or compliance documents required by the project, per-port optical report, serial/batch traceability and inspection method. For outside-plant use, define the enclosure and environmental scope as well as the optical chip.

Q: Should a new network deploy 1:128 GPON or XGS-PON?

A: If the goal is to serve many high-speed subscribers per port, XGS-PON capacity may be a better long-term answer than maximizing GPON sharing. The final choice depends on service tiers, installed equipment, coexistence plan and economics. Keep the passive ODN migration-ready whichever active technology is selected.

11. Conclusion: treat 1:128 as an engineered exception, not a default

The GPON TC-layer design considers 1:128, model-specific OLTs have advertised 128 clients per port, and commercial 1×128 PLC splitters exist. Those three facts do not approve a specific physical ODN. A reliable deployment is possible only when the worst optical path, upstream burst margin, busy-hour capacity, fault domain and future migration plan all pass together.

The most useful early check is simple: replace the theoretical 21.07 dB split with the splitter's guaranteed maximum, add every connector and splice, reserve a stated engineering allowance, and see what remains before fibre distance. In the transparent model used here, fixed loss is already 29.2 dB. That immediately rules out B+ and confines C+ to a short, well-controlled ODN.

For most FTTH networks, 1:32 or 1:64 remains easier to operate and upgrade. Use 1:128 where density, reach, take rate and economics genuinely justify it-and require the data to prove it. Glory Optical can support the passive side with a controlled PLC splitter and enclosure BOM, per-port test requirements and project drawings; final release should always be signed against the operator's OLT/ONU specifications and acceptance policy.

12. Standards and engineering references

• ITU-T G.984.1 - GPON general characteristics
• ITU-T G.984.2 (2019) - GPON PMD layer specification
• ITU-T G.984.3 (2014) - GPON transmission convergence layer specification
• Telcordia GR-1209-CORE - Generic requirements for passive optical components
• Telcordia GR-1221-CORE - Generic reliability assurance requirements
• Precision OT PLC splitter data sheet
• FS ABS PLC splitter data sheet
• SENKO passive optical splitter reliability white paper

For project application, pair these external references with Glory's ODN design guide, PLC splitter report guide and the limits stated on the approved project BOM.