Module 4
Optical Splitters
One light source. Many homes. The splitter is the small, powerless glass component that takes a single fiber's light and shares it among dozens of subscribers — and it does this by giving every home a smaller slice of the same beam. Understand the splitter and you understand why fiber networks are shaped the way they are.
What you'll be able to do
- Explain what a passive optical splitter does and why it needs no electricity.
- Read a split ratio (1:2 … 1:128) and say how many homes it serves.
- Tell single-stage from cascaded splits — and explain why cascading saves fiber but never saves light.
- Estimate splitter loss with the ~3-dB-per-doubling rule and read a real insertion-loss table.
- Choose between PLC (Planar Lightwave Circuit) and FBT (Fused Biconic Taper), and between balanced and unbalanced (tap) splitters.
The whole module at a glance: a passive splitter shares one beam many ways — and every share costs light.
The passive heart of the PON
A PON (Passive Optical Network) sends one signal out to many homes over shared glass. The component that makes that sharing possible is the optical splitter: one input fiber's light divided among N outputs. We write it as a ratio — 1:N — so a 1:32 has one fiber in, thirty-two out.
💡 The skylight analogy
Picture sunlight through one window. Put a prism in the beam and it fans into many smaller beams — no motor, no power. A splitter does exactly this for the laser light in a fiber: it fans one beam into many, passively.
Two facts make the splitter the quiet hero of fiber access:
- It has no power. "Passive" in PON means no electronics, no power supply, no cooling, nothing to fail in the field. A splitter is just precision glass in a sealed enclosure. That is why a fiber distribution cabinet can sit on a pole for twenty years with nothing plugged in.
- It is wavelength-agnostic. The splitter doesn't care what color of light passes through it — across its rated band it splits 1310 nm, 1490 nm, 1577 nm and the rest all the same way. (
nm= nanometre, the unit we use for a light's wavelength, i.e. its "color".)
🧭 Why this matters
Because the splitter treats every wavelength equally, the same physical ODN (Optical Distribution Network) — the fiber, splitters and connectors between operator and home — can carry every generation of PON at once. GPON, XGS-PON and future standards ride their own colors of light through the very same passive glass. What you bury today doesn't go obsolete when the electronics are upgraded.
1:N fan-out. One beam enters; N outputs leave. Each output carries the input power minus the splitter's loss — never more.
Split ratios and levels
Splitters come in binary ratios — each step doubles the outputs. The 1:32 is the most common distribution splitter in fiber-to-the-home today.
There are also 2:N variants (2:2 up to 2:64): a second input port, fed from two operator ports or two feeders. If one feeder is cut, traffic survives on the other — protection switching / redundancy.
Single-stage vs. cascaded splitting
The same total split can be reached two ways — single-stage (all splitting at one point, usually the FDH (Fiber Distribution Hub)) or cascaded (a primary split at the cabinet, then a secondary split in the field). The choice shapes the whole outside plant.
Cascade. Homes multiply (4 × 8 = 32), loss adds (≈ 7.3 + 10.75 ≈ 18.1 dB). Only one feeder leaves the cabinet — so you save feeder fiber, but not optical budget.
🧮 The multiplication rule
Total split = product of the stages. 1:4 × 1:8 = 1:32; 1:8 × 1:16 = 1:128. Many factor pairs reach the same total — what changes is where the second stage sits, not the home count.
⚠️ Cascading saves fiber, NOT light
Two small splitters are not "gentler" than one big one. In theory the loss of a cascade equals a single splitter of the same total ratio — 1:4 (6 dB) + 1:8 (9 dB) = 15 dB, exactly the same as one 1:32. In practice a cascade costs a touch more, because each stage adds its own excess loss plus an extra inter-stage connector pair — so expect a cascade to cost a dB or two more than the equivalent single splitter. Either way it's fiber economics, never optical-budget savings.
Every split costs you light
The single most important idea in the module: there is no free sharing. Divide power evenly among N outputs and each gets 1/N of the input. In decibels (dB) — the logarithmic unit for optical power — a clean pattern appears. (A related unit, dBm, measures absolute power relative to 1 mW: 0 dBm = 1 mW, and negative dBm means less than a milliwatt. dB is a ratio between two powers; dBm is a power.)
🧭 ~3 dB per doubling
Theoretical splitting loss of an ideal 1:N splitter is 10·log₁₀(N) dB. Doubling N adds 10·log₁₀(2) ≈ 3.01 dB, so every doubling loses about another 3 dB — half the light. Simple, relentless, and the reason fiber networks can't split forever.
The loss staircase. Bars show typical insertion loss (incl. excess loss), so the real steps run ~3–3.5 dB — a little more than the ideal +3 dB, and they grow with the ratio because excess loss accumulates. The 1:32 lands near 17.5 dB.
That formula is the ideal. A real splitter loses a bit more — manufacturing is imperfect, light scatters, and port connectors add their share. What you actually measure end to end is the insertion loss — the light a component "eats" just by being in the path:
real insertion loss = theoretical splitting loss + excess loss
Below are the numbers operators budget with: middle column is the clean math, right column is what a real splitter typically delivers (including excess loss).
| Split ratio (N) | Theoretical 10·log₁₀(N) | Typical insertion loss typical |
|---|---|---|
| 1:2 | 3.01 dB | ~3.5–4.2 dB |
| 1:4 | 6.02 dB | ~7.0–7.6 dB |
| 1:8 | 9.03 dB | ~10.5–11.0 dB |
| 1:16 | 12.04 dB | ~13.5–14.5 dB |
| 1:32 | 15.05 dB | ~17.0–17.8 dB |
| 1:64 | 18.06 dB | ~20.5–21.5 dB |
| 1:128 | 21.07 dB | ~23.5–26 dB |
⚠️ These numbers are genuinely noisy
Don't treat the right column as gospel — vendors and references disagree, depending on excess-loss assumptions and whether connectors are counted. For 1:32, most cite ~17 dB, but you'll see 16.5 dB and even 19 dB. For 1:128, modern chip-based datasheets reach ≤23.8 dB while the old rule of thumb says ~26 dB — use ~24–26 dB and label it approximate. A handy max-loss estimator: 0.8 + 3.4·log₂(N) dB.
The splitter dominates the loss budget
Why obsess over a few decibels? Because the operator's loss budget — the total light it can afford to lose between the operator's laser and the home — is small and fixed, and the splitter eats most of it. A standard GPON ITU-T Class B+ link has a 28 dB budget. A single 1:32 splitter (~17.5 dB) consumes well over half of that on its own.
That ~10.5 dB remainder must cover every kilometre of fiber and every splice and connector. Push to a 1:64 split (~21 dB) and there isn't enough left — which is why 1:64 generally needs a higher-power Class C+ (32 dB) operator port. The takeaway: the splitter is the dominant line item in the budget.
PLC vs. FBT
Splitters are built two ways, and the two suit different jobs: the PLC (Planar Lightwave Circuit) — waveguides etched onto a silica chip — and the FBT (Fused Biconic Taper) — fibers twisted, fused, and stretched together.
PLC — chip
- Waveguides etched on a silica chip
- Wide band ~1260–1650 nm, uniform
- Scales to 1:32, 1:64, 1:128
- Stable −40 to +85 °C
- Carrier default for FTTH
FBT — fused fibers
- Fibers twisted, fused, stretched
- Narrow band, loss varies by color
- Practical only to ~1:8
- Loss climbs when hot
- Low-ratio & tap splits
| Attribute | PLC (chip-based) | FBT (fused fibers) |
|---|---|---|
| How it's made | Waveguides etched on a silica chip | Fibers twisted, fused, stretched under heat |
| Wavelength range | Wide (~1260–1650 nm) — uniform across all PON colors | Narrow — tuned to 1–2 windows; loss varies by color |
| Practical max ratio | High — 1:32, 1:64, up to 1:128 | Low — practical to ~1:8 |
| Split uniformity | Excellent | Poorer at higher ratios |
| Temperature | −40 to +85 °C, stable | ~−5 to +75 °C; loss climbs when hot |
| Typical use | Backbone of GPON / XGS-PON networks | Low-ratio splits and unbalanced / tap ratios |
📌 Rule of thumb
PLC is the default for carrier fiber-to-the-home — its wide, uniform wavelength support is exactly what lets one network carry multiple PON generations, and it scales to high split counts with stable performance across temperature. FBT survives where the job is small or unusual: very low ratios, custom percentages, and the unbalanced taps we'll meet next.
Balanced vs. unbalanced (tap)
So far every output got an equal slice — a balanced splitter, which is what all the loss tables above assume (a 1:8 hands each port 12.5%). But sometimes you want an uneven split.
An unbalanced splitter (also asymmetric, tapered, or a tap — usually a 1:2) divides power by a chosen percentage: 1/99, 5/95, 10/90, 20/80. A small slice is "tapped" while most light continues downstream undisturbed.
| Type | Split | What it's for |
|---|---|---|
| Balanced | Even (e.g. 1:8 → 12.5% each) | Standard PON distribution — share equally among subscribers |
| Unbalanced / tap | Uneven (e.g. 10/90, 5/95) | Monitoring/test taps; "drop-as-you-go" linear builds; asymmetric protection paths |
🔎 Two places taps shine
Monitoring: a 5/95 tap pulls off ~5% of the light for a test instrument while leaving 95% for the customer — you watch the link without meaningfully disturbing it. Drop-as-you-go: along a rural or linear route, each access point taps a fixed percentage and passes the rest along the "bus," so homes near the start and far down the line end up with similar power. These are usually FBT (or custom PLC), because fused fibers naturally produce arbitrary split percentages.
Go deeper: the dB math behind the table optional
A decibel is a ratio on a logarithmic scale: loss in dB = 10·log₁₀(P_in / P_out). For an ideal even splitter each output gets 1/N of the input, so the loss is 10·log₁₀(N).
Two consequences worth internalizing:
- Doubling = +3 dB.
10·log₁₀(2) = 3.01, so each extra binary stage costs ~3 dB. That's why the theoretical column climbs in even ~3 dB steps. - Losses add in dB; powers multiply linearly. A cascade's stages each multiply the surviving fraction of power, and because dB is a log, multiplying fractions becomes adding dB. That is the mathematical reason cascading never saves optical budget —
1:4 (6 dB) + 1:8 (9 dB) = 15 dB, the very same theory as a single1:32.
Key takeaways
- A splitter passively divides one input fiber's light among N outputs (1:N) — no power, no electronics.
- It is wavelength-agnostic, which is exactly why one ODN carries every PON generation at once.
- Ratios are binary (1:2 … 1:128); each doubling adds outputs and ~3 dB of loss. 2:N adds a second input for redundancy.
- Single-stage centralizes the split; cascaded multiplies ratios (1:4 × 1:8 = 1:32) and saves feeder fiber — but losses add, so it never saves optical budget.
- Theoretical loss = 10·log₁₀(N); real insertion loss adds excess loss (1:32 ≈ 17.5 dB) and is genuinely noisy across vendors.
- The splitter dominates the loss budget (1:32 ≈ 17.5 dB of a 28 dB GPON B+ budget). PLC is the carrier default; balanced splits evenly, unbalanced taps unevenly.
Why can a single physical splitter carry GPON, XGS-PON, and future PON standards all at once?
Why: The splitter is passive and wavelength-agnostic across its rated band, so the same ODN can serve all PON generations simultaneously — that's the whole reason buried glass doesn't go obsolete.
Roughly how much loss does each doubling of the split ratio add, in theory?
Why: Each output gets 1/N of the power, so theoretical loss is 10·log₁₀(N) dB. Doubling N adds 10·log₁₀(2) ≈ 3.01 dB — about half the light per doubling.
A 1:4 primary feeds a 1:8 secondary. What does this cascade achieve?
Why: Cascaded ratios multiply (4 × 8 = 32 homes) and only one feeder fiber leaves the cabinet, saving fiber. But total loss = sum of the stages, so cascading never saves optical budget.
Why is PLC the default splitter technology for carrier fiber-to-the-home?
Why: PLC (chip-based) supports ~1260–1650 nm uniformly, scales to 1:32/1:64/1:128, and is stable from −40 to +85 °C. FBT is the one used for low-ratio and unbalanced/tap splits, not the carrier backbone.