This is what single-fibre working actually is, where the standards are, why the two ends differ, and what to check before you order.<\/p>\n\n\n\n
The first thing to settle: “BiDi” means two different products<\/h2>\n\n\n\n![\"BiDi](\"https:\/\/optics.carritech.com\/wp-content\/uploads\/2026\/08\/bidi-transceivers-two-kinds-1024x597.png\")
<\/figure>\n\n\n\nAsk two engineers what a BiDi transceiver is and you can easily get two answers, both correct.<\/p>\n\n\n\n
Single-fibre BiDi.<\/strong> 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.<\/p>\n\n\n\nData-centre “BiDi”.<\/strong> Two wavelengths in each direction over a duplex pair<\/strong> \u2014 still two strands, just carrying more per strand. IEEE’s 400GBASE-SR4.2, defined in Clause 150<\/a>, 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 \u2014 both directions on every fibre \u2014 was on the table and was not<\/strong> adopted; the group kept the usual convention of distinct transmit and receive fibres<\/a>.<\/p>\n\n\n\nSo 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.<\/p>\n\n\n\n
Where the BiDi transceiver standards actually are<\/h2>\n\n\n\n
This matters more than it sounds, because the naming in the market and the naming in the standard have drifted apart.<\/p>\n\n\n\n
At 100M and 1G, the IEEE clauses exist and are named BX.<\/strong> IEEE 802.3ah-2004 added Clause 58 for 100BASE-BX10 and Clause 59 for 1000BASE-BX10<\/a>, each defined in the standard as a point-to-point link “over one single mode fiber” \u2014 as against the duplex variants in the same clauses, which say “over two”. Ten kilometres, one strand, since 2004.<\/p>\n\n\n\nAt 10G, 25G and 50G, the IEEE name is BR, not BX.<\/strong> Single-fibre working above 1G was unstandardised for years \u2014 IEEE’s own call for interest in 2018<\/a> noted that the only bidirectional Ethernet access PHYs in existence came from that 2004 amendment. The gap was filled by IEEE 802.3cp-2021<\/a>, 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.<\/p>\n\n\n\nModules sold as “10GBASE-BX” are therefore using a market convention that predates the standard. That is not a criticism \u2014 the parts worked for a decade before IEEE described them \u2014 but if a specification says -BR<\/strong>, it is claiming an IEEE clause, and if it says -BX<\/strong> above 1G, it is not.<\/p>\n\n\n\nAbove 50G there is no IEEE single-fibre PMD at all.<\/strong> 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.<\/p>\n\n\n\nThe BiDi transceiver pairing rule, and why two of the same never link<\/h2>\n\n\n\n
This is the single most common ordering error with BiDi transceivers, and it is worth understanding rather than memorising.<\/p>\n\n\n\n
Each end transmits on one wavelength and listens on the other. IEEE 802.3ah defines downstream<\/strong> as transmission toward the subscriber end of a link and upstream<\/strong> as transmission away from it, which is where the D<\/strong> and U<\/strong> 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.<\/p>\n\n\n\nFit two D<\/strong> modules and both launch into the same band, and both have receivers filtered to the other<\/em> band. The far end’s light arrives \u2014 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.<\/p>\n\n\n\nPractical consequences worth building into how you order:<\/strong><\/p>\n\n\n\n\n- BiDi transceivers are ordered in pairs<\/strong>, and the pair is the unit. A spare is a pair, not a module.<\/li>\n\n\n\n
- Label both ends. A -U and a -D look identical in a drawer and identical in a cage.<\/li>\n\n\n\n
- Keep the pair together through a move. Half a pair is not a spare, it is a future fault.<\/li>\n\n\n\n
- Some naming conventions use A\/B, or the two wavelengths as a suffix, rather than U\/D. Same idea, different label \u2014 match the wavelengths, not the letters.<\/li>\n<\/ul>\n\n\n\n
Why the two ends are not symmetrical<\/h2>\n\n\n\n
There is a real optical reason the ends differ, beyond the wavelength assignment.<\/p>\n\n\n\n
Fibre attenuation is wavelength-dependent. ITU-T G.652<\/a>, 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\u20131565 nm band for G.652.D. Over a ten-kilometre span that is roughly a decibel of extra loss in one direction \u2014 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.<\/p>\n\n\n\nThe same asymmetry runs through the access standards. ITU-T G.984.2<\/a> specifies different launched-power ranges and different receiver sensitivities at the two ends of a GPON link, and defines separate path-loss classes.<\/p>\n\n\n\nReflection is the failure mode that surprises people<\/h2>\n\n\n\n
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 \u2014 and the reflected light comes straight back down the strand you are listening on.<\/p>\n\n\n\n
The Fiber Optic Association’s figures for reflectance by end-face type<\/a> are the numbers to know: a flat polish with an air gap reflects at about \u221220 dB, physical contact (PC) at \u221230 to \u221240 dB, ultra physical contact (UPC) at \u221240 to \u221250 dB, and angled physical contact (APC) at \u221260 dB or better.<\/p>\n\n\n\nThis is not theoretical. The FOA documents a campus single-mode network<\/a> 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.<\/p>\n\n\n\nThe 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.<\/p>\n\n\n\n
What follows from that:<\/strong><\/p>\n\n\n\n\n- Inspect and clean end-faces before you conclude a BiDi link is faulty. IEC 61300-3-35<\/a> is the standard that defines the inspection zones and the pass criteria \u2014 and it is explicit that an image alone cannot reject a connector; optical performance decides.<\/li>\n\n\n\n
- Where reflection is a known concern, APC is the standards-endorsed answer.<\/li>\n\n\n\n
- 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.<\/li>\n<\/ul>\n\n\n\n
Do not trust the diagnostics to two decimal places<\/h2>\n\n\n\n
Every modern module reports on itself \u2014 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.<\/p>\n\n\n\n
SFF-8472<\/a>, the SNIA-published management interface specification, states the accuracy goal for both transmit and receive optical power as better than \u00b13 dB<\/strong> over the specified temperature and voltage range. Not \u00b10.3 dB. Three.<\/p>\n\n\n\nTwo 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 \u2014 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<\/em>, 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.<\/p>\n\n\n\nSingle-fibre working is not exotic<\/h2>\n\n\n\n
It is easy to treat BiDi transceivers as a workaround for a fibre shortage. Structurally, they are the mainstream.<\/p>\n\n\n\n
Every mass-deployed fibre access architecture in the world is single-fibre bidirectional by design. GPON<\/a> runs downstream in the 1480\u20131500 nm band and upstream in the 1260\u20131360 nm band on one strand. XGS-PON<\/a> does the same at 9.95 Gbit\/s symmetric, downstream at 1575\u20131580 nm and upstream at 1260\u20131280 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.<\/p>\n\n\n\nHundreds 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.<\/p>\n\n\n\n
The strand arithmetic<\/h2>\n\n\n\n
Which brings us to why anyone buys them. Fibre counts, per link:<\/p>\n\n\n\n
<\/figure>\n\n\n\nAsk two engineers what a BiDi transceiver is and you can easily get two answers, both correct.<\/p>\n\n\n\n
Single-fibre BiDi.<\/strong> 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.<\/p>\n\n\n\n Data-centre “BiDi”.<\/strong> Two wavelengths in each direction over a duplex pair<\/strong> \u2014 still two strands, just carrying more per strand. IEEE’s 400GBASE-SR4.2, defined in Clause 150<\/a>, 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 \u2014 both directions on every fibre \u2014 was on the table and was not<\/strong> adopted; the group kept the usual convention of distinct transmit and receive fibres<\/a>.<\/p>\n\n\n\n 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.<\/p>\n\n\n\n This matters more than it sounds, because the naming in the market and the naming in the standard have drifted apart.<\/p>\n\n\n\n At 100M and 1G, the IEEE clauses exist and are named BX.<\/strong> IEEE 802.3ah-2004 added Clause 58 for 100BASE-BX10 and Clause 59 for 1000BASE-BX10<\/a>, each defined in the standard as a point-to-point link “over one single mode fiber” \u2014 as against the duplex variants in the same clauses, which say “over two”. Ten kilometres, one strand, since 2004.<\/p>\n\n\n\n At 10G, 25G and 50G, the IEEE name is BR, not BX.<\/strong> Single-fibre working above 1G was unstandardised for years \u2014 IEEE’s own call for interest in 2018<\/a> noted that the only bidirectional Ethernet access PHYs in existence came from that 2004 amendment. The gap was filled by IEEE 802.3cp-2021<\/a>, 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.<\/p>\n\n\n\n Modules sold as “10GBASE-BX” are therefore using a market convention that predates the standard. That is not a criticism \u2014 the parts worked for a decade before IEEE described them \u2014 but if a specification says -BR<\/strong>, it is claiming an IEEE clause, and if it says -BX<\/strong> above 1G, it is not.<\/p>\n\n\n\n Above 50G there is no IEEE single-fibre PMD at all.<\/strong> 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.<\/p>\n\n\n\n This is the single most common ordering error with BiDi transceivers, and it is worth understanding rather than memorising.<\/p>\n\n\n\n Each end transmits on one wavelength and listens on the other. IEEE 802.3ah defines downstream<\/strong> as transmission toward the subscriber end of a link and upstream<\/strong> as transmission away from it, which is where the D<\/strong> and U<\/strong> 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.<\/p>\n\n\n\n Fit two D<\/strong> modules and both launch into the same band, and both have receivers filtered to the other<\/em> band. The far end’s light arrives \u2014 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.<\/p>\n\n\n\n Practical consequences worth building into how you order:<\/strong><\/p>\n\n\n\n There is a real optical reason the ends differ, beyond the wavelength assignment.<\/p>\n\n\n\n Fibre attenuation is wavelength-dependent. ITU-T G.652<\/a>, 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\u20131565 nm band for G.652.D. Over a ten-kilometre span that is roughly a decibel of extra loss in one direction \u2014 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.<\/p>\n\n\n\n The same asymmetry runs through the access standards. ITU-T G.984.2<\/a> specifies different launched-power ranges and different receiver sensitivities at the two ends of a GPON link, and defines separate path-loss classes.<\/p>\n\n\n\n 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 \u2014 and the reflected light comes straight back down the strand you are listening on.<\/p>\n\n\n\n The Fiber Optic Association’s figures for reflectance by end-face type<\/a> are the numbers to know: a flat polish with an air gap reflects at about \u221220 dB, physical contact (PC) at \u221230 to \u221240 dB, ultra physical contact (UPC) at \u221240 to \u221250 dB, and angled physical contact (APC) at \u221260 dB or better.<\/p>\n\n\n\n This is not theoretical. The FOA documents a campus single-mode network<\/a> 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.<\/p>\n\n\n\n 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.<\/p>\n\n\n\n What follows from that:<\/strong><\/p>\n\n\n\n Every modern module reports on itself \u2014 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.<\/p>\n\n\n\n SFF-8472<\/a>, the SNIA-published management interface specification, states the accuracy goal for both transmit and receive optical power as better than \u00b13 dB<\/strong> over the specified temperature and voltage range. Not \u00b10.3 dB. Three.<\/p>\n\n\n\n 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 \u2014 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<\/em>, 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.<\/p>\n\n\n\n It is easy to treat BiDi transceivers as a workaround for a fibre shortage. Structurally, they are the mainstream.<\/p>\n\n\n\n Every mass-deployed fibre access architecture in the world is single-fibre bidirectional by design. GPON<\/a> runs downstream in the 1480\u20131500 nm band and upstream in the 1260\u20131360 nm band on one strand. XGS-PON<\/a> does the same at 9.95 Gbit\/s symmetric, downstream at 1575\u20131580 nm and upstream at 1260\u20131280 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.<\/p>\n\n\n\n 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.<\/p>\n\n\n\n Which brings us to why anyone buys them. Fibre counts, per link:<\/p>\n\n\n\nWhere the BiDi transceiver standards actually are<\/h2>\n\n\n\n
The BiDi transceiver pairing rule, and why two of the same never link<\/h2>\n\n\n\n
\n
Why the two ends are not symmetrical<\/h2>\n\n\n\n
Reflection is the failure mode that surprises people<\/h2>\n\n\n\n
\n
Do not trust the diagnostics to two decimal places<\/h2>\n\n\n\n
Single-fibre working is not exotic<\/h2>\n\n\n\n
The strand arithmetic<\/h2>\n\n\n\n