Where 100G backhaul sits in a 5G network
A radio site terminates its fronthaul links locally and sends the aggregated traffic back towards the core. That is the backhaul; in dense areas the same span also carries mid-haul traffic between distributed and centralised units. One aggregation node can collect dozens of radio sites, so the backhaul port carries the sum of them on one link, normally in a ring so that a single cut does not isolate a site.
100G is the working rate for that link. 25G sits at the site edge and on the fronthaul ports, 100G aggregates a group of sites, and 400G sits at the top of the hierarchy. Choosing the 100G optic is mostly a span question - distance, loss, fibre count, whether a line system exists - and only secondarily a question about the module.
Why 100G QSFP28 is the backhaul default
QSFP28 delivers 100G in a cage long enough in service to be well understood, with several interface types behind the same footprint: SR4 and CWDM4 for short campus and aggregation runs, PSM4 for parallel-fibre sites, LR4 for a standard 10 km span, ER4 for a longer one, and a coherent ZR part where an amplified design makes more sense than an extra regenerator. The same port can break out to four 25G links where the far end is a group of radio sites rather than one node.
Recommended optics for 100G backhaul
| Module | Rate | Media / wavelength | Reach | Typical use |
|---|
| 100G QSFP28 SR4 | 100G | MMF, 850 nm parallel | 70 to 100 m | Within a hub or shelter, patching between aggregation frames |
| 100G QSFP28 CWDM4 | 100G | SMF, CWDM 1271 to 1331 nm | 2 km | Campus and short metro runs between aggregation sites |
| 100G QSFP28 PSM4 | 100G | SMF, 1310 nm parallel | 500 m to 2 km | Parallel-fibre plants and short runs to a data hall |
| 100G QSFP28 LR4 | 100G | SMF, LAN-WDM 1295 to 1310 nm | 10 km | Standard backhaul span in a metro ring |
| 100G QSFP28 ER4 | 100G | SMF, LAN-WDM | 30 to 40 km | Longer backhaul span or sparse site layout |
| 100G QSFP28 ZR | 100G coherent | DWDM C-band, tunable | 80 km class and beyond with amplification | Amplified backhaul, DWDM ring, muxponder uplink |
| 100G QSFP28 BiDi | 100G | SMF, single-fibre wavelength pair | 10 to 20 km | Fibre-poor sites where only one strand is free |
| 100G to 4x25G breakout | 100G host, 4x25G links | SMF or MMF plus breakout assembly | Span dependent | Splitting an aggregate port towards four radio sites |
Reach figures are the standard classes for each interface type; the working limit is the loss budget. The full set of form factors and distances is on the 100G QSFP28 transceiver page, and the site-edge ports that feed this link are covered by the 25G SFP28 family.
CWDM, DWDM or dark fibre
Backhaul rings are usually built on fibre that is already there. Three answers are common. On dark fibre, one LR4 or ER4 pair per 100G link is the simplest design and scales by adding pairs. On CWDM or DWDM with passive muxes, several 100G links share one fibre pair, which is what most metro rings use once the pair count runs out. For long amplified spans, a coherent 100G ZR channel on the DWDM grid carries traffic further without a regenerator. Decide this first: it sets the wavelength plan the modules are ordered against.
Timing and synchronisation across the backhaul
Backhaul and mid-haul links carry PTP and SyncE alongside the user plane, and the optical layer passes that timing without adding jitter. Two practical points follow. Latency and FEC class should be consistent along a chain, because mixing generations on one path changes the delay budget and complicates fault-finding. And on a link close to its reach limit, a longer-reach module beats an extra amplifier, since every added element adds delay and a failure point.
Pole, cabinet and shelter temperature
Aggregation cabinets and pole-mounted equipment see the full outdoor range, and the module inside is often the first component to drift. Commercial grade optics rated for a conditioned room are a common cause of a link that works at commissioning and flaps in midsummer. Specify the industrial grade for anything outside a cooled shelter, state it in the RFQ so it is not substituted, and keep the grade consistent around a ring so the fault pattern stays predictable.
Coding, breakout and compatibility
The module has to be coded for the host you name, and a breakout assembly has to match the host's port configuration where a 100G port is split into four 25G links. Modules supplied for branded platforms are compatible parts and are not manufactured by the brand of the host, with no claim of relationship to that brand. Name the platform model and port type in the order, and note whether the link has to interoperate with optics already in service at the far end. Vendors seen in this layer include Cisco, Nokia, Huawei and ZTE platforms, among others.
How to choose, in order
- Fix the span first: distance, measured loss, and whether amplifiers or a line system are in the path.
- Decide the transport: dark fibre per link, CWDM or DWDM sharing a pair, or coherent on the DWDM grid.
- Set the interface type to the distance class: SR4, CWDM4, PSM4, LR4, ER4 or ZR.
- Check whether the far end is one node or four radio sites, and choose native 100G or 4x25G breakout.
- Set the temperature grade from where the module sits, and keep it consistent around the ring.
- Confirm timing and latency, then name the host model and port type for coding and spares.
Frequently asked questions
Is 100G enough for 5G backhaul?
For aggregating a group of radio sites it is the working rate today, with 400G at the upper layers. Where one span carries an unusually large cluster, plan the 400G upgrade for that span.
LR4 or ER4 for a 25 km span?
ER4 is the usual choice. LR4 is specified to the 10 km class, and running beyond that spends margin that should be kept for ageing and repairs.
When is a coherent 100G ZR module the better answer?
On amplified spans and DWDM rings, and on long links where a regenerator or amplifier site would otherwise be needed. On a short unamplified span, LR4 or ER4 is simpler and cheaper.
Do the backhaul optics have to pass PTP and SyncE?
Yes, and the optical layer passes timing without adding jitter. Keep latency and FEC class consistent along the chain, and confirm the requirement with the transport and RAN vendors.
Can one 100G port feed four radio sites?
Yes, with a breakout to four 25G links where the host port supports it and the far ends are 25G. Confirm the breakout assembly and host configuration together, since coding and port mapping both have to match.
Send the span distance, the measured loss, the host platform and the number of sites behind the link, and we confirm the interface type, temperature grade and coding before shipment. Related reading: WDM transceivers for the CWDM and DWDM options, 400G QSFP-DD and OSFP optics for the layer above, and all optics selection guides.
