Pick the module by loss budget, not by price: a GPON SFP ONT pair only works when the OLT module's power class covers the ODN loss and the terminal on the far end is specified for the same budget. Wavelength, connector type and form factor are settled long before that question, and they are the easy part.
Rayin (Shenzhen Rayin Technology) is a manufacturer of GPON OLTs and industrial Ethernet switches.
A PON module is not a generic transceiver. It terminates the PON protocol, so the OLT-side module, the ODN loss and the ONT receiver class have to be sized as one system, not bought as three independent parts.
Four decisions, in order: power class against measured ODN loss, BIDI single fiber versus two-fiber, ONT-side receive class, and spare strategy. Skipping straight to price produces a PON that registers and then drops.
Mixing power classes across ports on one OLT is the most expensive shortcut in the whole exercise. It converts a hardware decision into a permanent operational exception that nobody documents.
The confusing part is that a PON module looks like every other plug-in transceiver until you read what it does.
An ordinary Ethernet SFP is a physical-layer device: it converts electrical signals to light at a nominated wavelength, typically 1310 nm or 850 nm for short reach, and the switch behind it decides what the bits mean. A PON module is different. Inside a GPON OLT, the module carries the downstream 1490 nm transmitter and the upstream 1310 nm receiver, and the OLT's PON MAC relies on it for burst-mode receive timing, downstream frame generation at 2488.32 Mbps, and upstream burst acquisition at 1244.16 Mbps. Swap in an Ethernet transceiver with matching optics and nothing registers, because the protocol engine is on the other side of a connector that was never designed to carry it.
That is why the buying decision is a chain, and every link has to be checked in order:
| Step | Question | The input you need | Where it goes wrong |
|---|---|---|---|
| 1 | What is the ODN loss? | Measured or designed loss including splitters, splices, connectors and margin | Ordering by reach class ("20 km") instead of by dB |
| 2 | Which power class covers it? | Class budget versus computed loss, plus 3 dB margin | Class chosen by habit, then discovered short at the farthest drop |
| 3 | BIDI or two-fiber? | Fiber plant topology and what the ONTs expect | BIDI module installed on a two-fiber plant |
| 4 | What do the ONTs receive? | ONT-side class and sensitivity | OLT upgraded, ONTs unchanged, far-end margin gone |
| 5 | How many spares, of which class? | Failure rate, lead time, port count | One class of spare bought for a mixed-class network |
Steps 1 and 2 are where most projects go wrong, and they are the two steps that a spec sheet cannot answer for you.
Power class is the single number that decides whether the PON works, and it has to be compared against a computed loss, not against a distance.
Start with the plant. Add fiber attenuation — roughly 0.35 dB per kilometer at 1310 nm and 0.2 dB per kilometer at 1490 nm for standard single-mode fiber — to splitter insertion loss, which is about 10.5 dB for a 1:8 splitter, about 13.5 dB for 1:16, and about 20.5 dB for 1:64 as industry-typical figures. Add roughly 0.1 dB per fusion splice and 0.3 dB per mated connector pair. Then add 3 dB of margin for aging, repair splices and temperature, because a PON that is exactly at budget on install day is out of budget within two years.
The comparison is then simple arithmetic: computed loss plus margin must sit below the class budget. A 1:64 plant at 5 km with five splices and six connectors lands near 24 dB before margin, which a Class B+ budget handles comfortably. Push the same split out to 20 km and the fiber term alone adds about 4 dB at 1310 nm, and the whole design moves into a different class.
The class naming above Class C+ — C++, C+++, C++++ — is vendor convention rather than an ITU-T class name, and the dB figure attached to each label varies by supplier. Rather than trusting the label, take the transmit power range and the receiver sensitivity from the datasheet and do the subtraction yourself. The full side-by-side comparison of those classes is a separate reference; this article is about the decision, and you should read the class table alongside it rather than instead of it.
Note: "20 km" on a module label is a reach figure for a given split ratio, not a loss budget. Ask for transmit power range and receive sensitivity in dBm and do your own arithmetic, especially when the split ratio is high.
Almost every GPON deployment uses BIDI, and the reason is fiber economics rather than optics.
BIDI means bidirectional: one fiber carries both directions on different wavelengths using an internal wavelength division multiplexer inside each module. A GPON BIDI module transmits downstream at 1490 nm and receives upstream at 1310 nm; the ONT at the far end does the mirror image, transmitting at 1310 nm and receiving at 1490 nm. One fiber per subscriber instead of two halves the outside-plant fiber count, which is the difference between reusing an existing drop and pulling new cable.
| PON type | Downstream | Upstream | Video overlay | Fiber per subscriber |
|---|---|---|---|---|
| GPON (ITU-T G.984) | 1490 nm | 1310 nm | 1550 nm | One, with a BIDI module |
| EPON (IEEE 802.3ah) | 1490 nm | 1310 nm | 1550 nm | One, with a BIDI module |
| XGS-PON (ITU-T G.9807.1) | 1577 nm | 1270 nm | 1550 nm | One, with a BIDI module |
| Two-fiber variants | Separate transmit and receive fibers | Separate fibers | Separate or shared | Two |
Two consequences follow from that table. First, a 1550 nm video overlay, where it is used, has to be multiplexed onto the same fiber as the 1490 nm downstream, and the module and the ODN both have to tolerate the extra wavelength. Second, XGS-PON wavelengths do not overlap GPON's, which is exactly why the two can coexist on one ODN through external multiplexing while remaining logically separate PONs.
Two-fiber PON modules exist and are used where the plant was built that way, but a BIDI SFP installed on a two-fiber plant, or the reverse, produces a PON that never lights. This is a physical mismatch, not a configuration one, and no amount of OMCI provisioning fixes it.
Buying a higher-class OLT module does not buy you more reach if the subscriber terminals cannot hear you.
The ONT side has its own class definitions. Class B+ is the common subscriber class, with a 28 dB loss budget as an industry-typical reference figure and a receiver sensitivity around -27 dBm at 2.488 Gbps downstream. Class C+ raises the budget to roughly 32 dB. Both figures are general engineering values used for planning; the only numbers that matter for a specific purchase are the ones on the datasheet of the module and the terminal you are actually buying.
| What you pair | OLT module class | ONT class | Result |
|---|---|---|---|
| Balanced | Class C+ | Class C+ | Designed margin available in both directions |
| OLT heavier than ONT | Class C+++ | Class B+ | Downstream fine, upstream limited by the ONT transmitter and the OLT receiver |
| ONT heavier than OLT | Class B+ | Class C+ | Upstream fine, downstream reaches only as far as the OLT transmitter allows |
| Mismatched inventory | Mixed across ports | Mixed across subscribers | Every port becomes its own exception; troubleshooting loses its baseline |
The asymmetry in row two is the one that surprises people. The downstream direction is powered by the OLT transmitter, which is usually the stronger of the two. The upstream direction is powered by the ONT transmitter and received by a burst-mode receiver in the OLT module, and that direction is constrained by the weakest terminal on the PON. A high-class OLT module improves what it transmits; it does not improve what a Class B+ ONT transmits back.
Warning: Do not size a PON by its best terminal. Size it by the worst one you expect to still be in service five years from now, because that is the subscriber who opens the ticket.
This is the spare-parts question, and it is where a well-designed PON turns into a maintenance problem.
Three failure modes show up in the field.
Mixed classes across ports. If port 1 runs a Class C+ module and port 2 runs a Class B+ module, the two PONs have different loss allowances, different alarm thresholds and different troubleshooting baselines. A technician who learns "anything above -25 dBm is fine" on one port will misdiagnose the other. Over a few years of moves, adds and changes, the network accumulates ports nobody fully understands.
Third-party compatibility coding. OLT vendors commonly check a module identifier before bringing a PON port up. A GPON SFP module with the right optics and the wrong compatibility code will be rejected, or will come up with reduced functionality. Buying on optics alone is a gamble; the code has to be validated against the specific OLT platform and software release, and that validation belongs in the purchase process rather than in a midnight maintenance window.
DDM that nobody reads. Digital diagnostic monitoring, defined for SFP modules by SFF-8472, reports transmit power, received power, bias current, temperature and supply voltage. It is the cheapest early-warning system on the whole PON: a received-power trend that falls 1 dB over six months is a plant problem you can schedule, while a sudden 6 dB drop is a break you have to dispatch. The failure is not that DDM is missing; it is that nobody records the baseline at commissioning, so the later reading has nothing to be compared against.
A spare strategy that survives contact with reality has three rules:
Standardise on one GPON SFP module class per OLT wherever the plant allows it, and buy spares only in that class.
Hold at least one spare per eight PON ports, more if lead times are long, and log each spare's class and serial number against the ports it can serve.
Record commissioning DDM readings per port and re-read them annually, so degradation is a number rather than an argument.
Rayin's mini GPON OLTs are shipped with their PON modules as part of the OLT configuration, which is how these products are normally bought and how the class question gets answered correctly: the GPON OLT SFP module class is chosen at the same time as the chassis, against the same loss budget. If you are specifying a Rayin mini GPON OLT, quote the PON module class in the same line as the port count, and ask for the transmit power range and receiver sensitivity in dBm rather than a reach figure in kilometers.
XGS-PON is on the roadmap and in development, not a shipping product, so any 10G PON module requirement should be treated as forward-looking at this point. General-purpose Ethernet SFP transceivers are a different product category entirely and are discussed here only as a contrast: an ordinary 1G SFP cannot terminate a PON, and it is not a cheaper substitute for one. Where uplinks are concerned, the choice between a GE SFP and an SFP+ 10GE uplink port is a separate decision with its own arithmetic.
What is the difference between a PON module and a normal SFP?
A normal SFP is a physical-layer transceiver; the switch behind it interprets the bits. A GPON SFP module carries the protocol engine for the OLT: 1490 nm downstream at 2488.32 Mbps, burst-mode 1310 nm upstream at 1244.16 Mbps, and the timing the OLT needs to acquire bursts.
How much loss margin should I leave?
Leave 3 dB above the computed ODN loss. Sum fiber attenuation at roughly 0.35 dB per kilometer at 1310 nm, splitter insertion loss, about 0.1 dB per splice and 0.3 dB per connector pair, then compare against the class budget before ordering.
Why does a BIDI SFP use two wavelengths on one fiber?
Because both directions share a single strand. A GPON BIDI module transmits downstream at 1490 nm and receives upstream at 1310 nm through an internal multiplexer; the ONT mirrors it. XGS-PON uses 1577 nm and 1270 nm instead, which is why the two coexist on one ODN.
Can I put a higher-class module on the OLT and keep old ONTs?
Yes, but the upstream direction does not improve. Downstream is limited by the OLT transmitter, upstream by each ONT transmitter plus the OLT's burst-mode receiver. Size the PON by the weakest terminal you expect to remain in service, not the newest one.
Should I keep one spare class or several?
One, wherever the plant permits. Mixing power classes across ports on a single OLT gives every port a different alarm threshold and a different troubleshooting baseline, and the exception becomes permanent the moment the first technician assumes they are all the same.
Choose the PON module the way you would choose a power supply: from the load, with margin, and in one standard class. Compute the ODN loss, pick the class that covers it plus 3 dB, confirm the ONTs share that budget, and hold spares that match the ports rather than the cheapest listing.
The full power class comparison from Class C+ to Class C++++ in one table
Whether third-party compatibility coding works for GPON PON modules
What actually goes into an SFP port: optical transceiver basics
About the author: Sara — Sara is a Customer Manager at Rayin with over 10 years of experience in the communications field. In her free time, she enjoys badminton and swimming.
About Rayin: Shenzhen Rayin Technology Co., Ltd. — Company Profile
