BiDi transceivers: one fibre, two wavelengths, and the pairing rule that catches everyone

BiDi transceivers do something that sounds like it should not work: they carry a full-duplex link over a single strand of fibre. One strand, both directions, at the same time.

They are also the source of more confused ordering than any other part of the optics catalogue — because the word “BiDi” is used in the industry for two entirely different products, and because a BiDi link needs two different modules that most part numbers make look identical.

This is what single-fibre working actually is, where the standards are, why the two ends differ, and what to check before you order.

The first thing to settle: “BiDi” means two different products

Ask two engineers what a BiDi transceiver is and you can easily get two answers, both correct.

Single-fibre BiDi. One strand carries both directions on two different wavelengths. The module contains a wavelength-selective filter that launches its own transmit wavelength into the fibre and passes only the complementary band inward to its receiver. This is what BX and BR parts are, and it is what the rest of this article is about.

Data-centre “BiDi”. Two wavelengths in each direction over a duplex pair — still two strands, just carrying more per strand. IEEE’s 400GBASE-SR4.2, defined in Clause 150, is the standardised example: eight multimode fibres, two wavelengths per fibre, four fibres per direction. When IEEE’s own task force set the baseline for it, a genuinely bidirectional option — both directions on every fibre — was on the table and was not adopted; the group kept the usual convention of distinct transmit and receive fibres.

So a part described as “BiDi” may halve your strand count or may not change it at all. If the fibre saving is the reason you are buying, that is the first question to ask.

Where the BiDi transceiver standards actually are

This matters more than it sounds, because the naming in the market and the naming in the standard have drifted apart.

At 100M and 1G, the IEEE clauses exist and are named BX. IEEE 802.3ah-2004 added Clause 58 for 100BASE-BX10 and Clause 59 for 1000BASE-BX10, each defined in the standard as a point-to-point link “over one single mode fiber” — as against the duplex variants in the same clauses, which say “over two”. Ten kilometres, one strand, since 2004.

At 10G, 25G and 50G, the IEEE name is BR, not BX. Single-fibre working above 1G was unstandardised for years — IEEE’s own call for interest in 2018 noted that the only bidirectional Ethernet access PHYs in existence came from that 2004 amendment. The gap was filled by IEEE 802.3cp-2021, which adds Physical Layer specifications for “10 Gb/s, 25 Gb/s, and 50 Gb/s Ethernet optical interfaces for bidirectional operation over a single strand of single-mode fiber with reaches of at least 10 km, 20 km, and 40 km”. The PMD names are 10GBASE-BR10/BR20/BR40 and the same at 25G and 50G.

Modules sold as “10GBASE-BX” are therefore using a market convention that predates the standard. That is not a criticism — the parts worked for a decade before IEEE described them — but if a specification says -BR, it is claiming an IEEE clause, and if it says -BX above 1G, it is not.

Above 50G there is no IEEE single-fibre PMD at all. 802.3cp stops at 50 Gb/s. Single-strand 100G links exist and are widely deployed, but they sit outside IEEE 802.3 and are matched pair to pair rather than by clause. Worth knowing before someone asks you which standard your 100G BiDi link complies with.

The BiDi transceiver pairing rule, and why two of the same never link

This is the single most common ordering error with BiDi transceivers, and it is worth understanding rather than memorising.

Each end transmits on one wavelength and listens on the other. IEEE 802.3ah defines downstream as transmission toward the subscriber end of a link and upstream as transmission away from it, which is where the D and U suffixes come from. At 1G the downstream end transmits at around 1490 nm and receives in the 1310 nm band; the upstream end does the reverse.

Fit two D modules and both launch into the same band, and both have receivers filtered to the other band. The far end’s light arrives — and is rejected by the near end’s own filter before it reaches the detector. The link never establishes. Diagnostics make it look baffling: both ends report healthy transmit power and neither reports any meaningful receive power, which reads like a broken fibre rather than a mismatched pair.

Practical consequences worth building into how you order:

• BiDi transceivers are ordered in pairs, and the pair is the unit. A spare is a pair, not a module.
• Label both ends. A -U and a -D look identical in a drawer and identical in a cage.
• Keep the pair together through a move. Half a pair is not a spare, it is a future fault.
• Some naming conventions use A/B, or the two wavelengths as a suffix, rather than U/D. Same idea, different label — match the wavelengths, not the letters.

Why the two ends are not symmetrical

There is a real optical reason the ends differ, beyond the wavelength assignment.

Fibre attenuation is wavelength-dependent. ITU-T G.652, which specifies the single-mode fibre almost everyone is running, sets a maximum of 0.40 dB/km at 1310 nm against 0.30 dB/km in the 1530–1565 nm band for G.652.D. Over a ten-kilometre span that is roughly a decibel of extra loss in one direction — so the two directions cannot share one power budget, and the standards do not ask them to. The IEEE clauses specify the two ends separately, with different receive sensitivities at each end.

The same asymmetry runs through the access standards. ITU-T G.984.2 specifies different launched-power ranges and different receiver sensitivities at the two ends of a GPON link, and defines separate path-loss classes.

Reflection is the failure mode that surprises people

On a duplex link, a poor connector end-face costs you loss. On a single-fibre link, that same end-face is a mirror sitting a few metres from a transmitter that is tens of decibels stronger than the signal you are trying to receive — and the reflected light comes straight back down the strand you are listening on.

The Fiber Optic Association’s figures for reflectance by end-face type are the numbers to know: a flat polish with an air gap reflects at about −20 dB, physical contact (PC) at −30 to −40 dB, ultra physical contact (UPC) at −40 to −50 dB, and angled physical contact (APC) at −60 dB or better.

This is not theoretical. The FOA documents a campus single-mode network with a high bit error rate whose loss measurements were all fine; the cause was an optical return loss of only 37 dB across a handful of connections, producing multipath interference. Replacing the connectors with APC resolved it.

The standards bodies take the same view. ITU-T’s GPON specification sets a normative floor of more than 32 dB optical return loss on the distribution network, in both directions, precisely because single-fibre working makes reflection a first-order concern rather than a second-order one.

What follows from that:

• Inspect and clean end-faces before you conclude a BiDi link is faulty. IEC 61300-3-35 is the standard that defines the inspection zones and the pass criteria — and it is explicit that an image alone cannot reject a connector; optical performance decides.
• Where reflection is a known concern, APC is the standards-endorsed answer.
• A high bit error rate with a healthy loss measurement is the classic signature. Reach for an OTDR to find the reflective event rather than re-measuring loss.

Do not trust the diagnostics to two decimal places

Every modern module reports on itself — transmit power, receive power, bias current, temperature, supply voltage. On a BiDi link that readout is the main diagnostic you have, so it is worth knowing how accurate it is.

SFF-8472, the SNIA-published management interface specification, states the accuracy goal for both transmit and receive optical power as better than ±3 dB over the specified temperature and voltage range. Not ±0.3 dB. Three.

Two things follow. Read the two ends of a link and the round-trip uncertainty can be six decibels, which is more than the margin on plenty of real links — so diagnostics are a tool for spotting gross faults, not for proving a power budget. And the receive accuracy is specified at the vendor-specified wavelength, which on a BiDi module is the far end’s transmit wavelength. The calibration is inherently pair-specific, which is one more reason the pair is the unit.

Single-fibre working is not exotic

It is easy to treat BiDi transceivers as a workaround for a fibre shortage. Structurally, they are the mainstream.

Every mass-deployed fibre access architecture in the world is single-fibre bidirectional by design. GPON runs downstream in the 1480–1500 nm band and upstream in the 1260–1360 nm band on one strand. XGS-PON does the same at 9.95 Gbit/s symmetric, downstream at 1575–1580 nm and upstream at 1260–1280 nm. EPON is single-fibre, standardised in the same 2004 amendment that gave us the BX clauses. NG-PON2 is single-fibre with tunable wavelengths.

Hundreds of millions of subscriber links run this way. The technique is not unproven; it is simply less common in the enterprise, where fibre has historically been cheap enough not to think about.

The strand arithmetic

Which brings us to why anyone buys them. Fibre counts, per link:

The headline is the last row. 100GBASE-SR4 is parallel transmission over eight multimode fibres on an MPO connector. Against a two-strand single-lambda 100G part that is a four-to-one difference in fibre, and against a single-fibre link it is eight to one.

On a new build where you are pulling fibre anyway, that is a cost line. On a retrofit where the plant is already in the ground and the ducts are full, it is frequently the whole business case — the difference between an upgrade you can do this quarter and a civils project you cannot.

What to check before you order BiDi transceivers

1. Which BiDi you mean. Single-strand, or a duplex pair carrying two wavelengths each way? They are different products with different fibre savings.
2. The reach class, honestly measured. Single-fibre parts are specified per reach class the same way duplex parts are, and fitting an 80 km part on a 200 m link risks overloading the receiver. Too much power is a fault, not a safety margin.
3. The pair, not the module. Order and stock in pairs, label both ends, and keep them together.
4. The end-faces. Inspect and clean before commissioning, and consider APC where reflection is a known concern.
5. What the far end already is. If you are adding to an existing single-fibre link, you need the complementary end at the matching wavelengths and reach class — which is a lookup, not a guess.

Check a part, or send us the list

Our single-fibre range runs from 155M to 100G, in reach classes from 3 km to 180 km, with both ends of every pair — including the dual-channel CSFP variants that put two single-fibre links in one cage.

If you have a part number in front of you, the compatibility checker will tell you what it maps to in about ten seconds, by part number or by hardware model.

If you have a whole list — a bill of materials, someone else’s quote, a spreadsheet — send it over and we will come back line by line with the compatible equivalent and the specification against each one. On BiDi links, tell us which end you need, or tell us both and we will pair them for you.