Where mid-haul sits in a 5G transport network
Mid-haul is the link between the distributed unit at or near the radio site and the centralised unit further back. It is not a fronthaul link: the traffic is already packetised and aggregated, distances are longer, and the span usually runs between two aggregation points rather than between two units in one shelter. One mid-haul link can collect the traffic of a group of sites, so the port rate is set by how many sites sit behind it and how much of that traffic is latency-sensitive.
50G is the rate most often used for that layer today, with 25G at the site edge and 100G where a link aggregates several groups or begins to carry data centre traffic. Choosing the optic is mostly a question about span, ring layout and temperature; only then about the module.
Why mid-haul landed on 50G
50G gives a clean step above the 25G ports that dominate the radio edge without the cost and power of a 100G port on every mid-haul span. Two form factors matter: 50G SFP56, which keeps the SFP-family cage and one lane of PAM4, and 50G QSFP28, which reaches the same rate with two 25G lanes and suits hosts already built around QSFP cages. Where the span carries aggregated traffic from many sites, or the operator is planning to consolidate centralised units, 100G QSFP28 with LR4 or ER4 optics is the usual upgrade path and shares the same fibre.
Recommended optics for 5G mid-haul
| Module | Rate | Media / wavelength | Reach | Typical mid-haul use |
|---|
| 50G SFP56 SR | 50G | MMF, 850 nm | 70 to 100 m | Patch between aggregation frames in the same hub |
| 50G SFP56 LR | 50G | SMF, 1310 nm | 10 km | Standard mid-haul span in a metro ring |
| 50G SFP56 ER | 50G | SMF, 1310 nm | 30 to 40 km | Longer mid-haul span, sparse aggregation layout |
| 50G SFP56 BiDi | 50G | SMF, single-strand wavelength pair | 10 to 20 km | Fibre-poor routes where only one strand is free |
| 50G SFP56 CWDM / DWDM | 50G | SMF, CWDM or DWDM channels | 10 to 20 km | Several 50G links sharing one fibre pair |
| 50G QSFP28 (2 x 25G) | 50G | SMF, 1310 nm or CWDM | 2 to 10 km | Hosts built around QSFP cages rather than SFP56 |
| 100G QSFP28 LR4 / ER4 | 100G | SMF, LAN-WDM | 10 to 40 km | Upgrade path where several site groups are aggregated |
| 25G SFP28 LR / ER | 25G | SMF, 1310 nm | 10 to 40 km | Site-edge uplinks feeding the mid-haul node |
Reach figures are standard classes; the working limit is the measured loss on the span. The 50G options are listed on the 50G SFP56 transceiver page, the 25G site-edge ports on the 25G SFP28 family, and the aggregation layer above on the 100G QSFP28 page.
Reach, ring layout and protection
Mid-haul spans are usually built as rings so that one cut does not isolate a group of sites. That affects the optics in three ways. The span loss of the longest path around the ring decides the reach class, not the average. Protection switching means both paths have to come up with the same module type and the same FEC setting, otherwise the working path looks healthy while the protected path never closes at margin. And where a ring is extended, the extra span is best covered by a longer-reach part rather than by inserting an amplifier into a ring that was not planned for one.
Timing and latency across the mid-haul
Mid-haul links carry PTP and SyncE along with user traffic, and they are the layer where the delay budget of a radio site is largely set. Keep the latency class and FEC profile consistent across a chain of links, because a single mixed-generation hop changes the budget and complicates fault-finding.
Temperature, power and cabinet limits
Mid-haul equipment lives in aggregation cabinets, rooftop enclosures and small huts, and a module is often the first component to drift when a cabinet runs hot. Specify the industrial grade for anything outside a conditioned room, and state it in the order so it is not substituted. Power is the second cabinet limit: a 50G SFP56 port draws less than a 100G port, which is a real constraint when the cabinet has no spare capacity, so check the port power budget before filling a line card.
Coding, compatibility and spares
The module has to be coded for the host you name and tested against the port. 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 the port type, and say whether the link has to interoperate with optics already in service at the far end. Platforms seen at this layer include Nokia, Huawei, ZTE and Cisco. Keep one spare of each link type on site: a failed mid-haul optic takes a whole group of sites off the air.
How to choose, in order
- Set the rate from the traffic behind the link: 50G for a group of sites, 100G where groups are aggregated.
- Measure the longest path around the ring and choose LR, ER or a WDM channel against that loss.
- Decide the fibre plan: dual strand per link, BiDi where one strand is free, CWDM or DWDM where a pair is shared.
- Match the form factor to the host cage, SFP56 or QSFP28, and check the port power budget.
- State the temperature grade and the timing requirement with the order.
- Name the host model for coding, and keep spares of each type on site.
Frequently asked questions
Is mid-haul always 50G?
No. 50G is the common rate for a group of sites, but 25G is still used on short or lightly loaded spans and 100G where a link aggregates several groups or carries data centre traffic.
50G SFP56 or 50G QSFP28?
Both deliver 50G. SFP56 keeps one lane in an SFP-family cage and is the smaller, lower-power option; QSFP28 suits hosts built around QSFP cages. The host port decides.
LR or ER for a mid-haul span?
LR covers the 10 km class and ER the 30 to 40 km class. Choose against measured span loss on the longest protected path, not against the drawing.
Does mid-haul need the industrial temperature grade?
Yes, wherever the equipment sits in a cabinet, rooftop enclosure or hut rather than a conditioned room. State the grade in the order so a commercial part is not substituted.
Can several mid-haul links share one fibre pair?
Yes, with CWDM or DWDM modules and a passive mux. Both ends of a channel are ordered as a matched pair, and the wavelength plan is set before the modules are chosen.
Send the host platform and port type, the span distance and measured loss, the temperature range and the ring layout, and we confirm the rate, the reach class and the coding before shipment. Related reading: optics for 5G backhaul for the layer above, WDM transceivers for shared-fibre routes, and all optics selection guides.
