transceiver-db/blog-training-data/blog-090-optics-for-5g-fronthaul-midhaul.md

title	slug	type	category	tags	seo_focus_keyword
Optics for 5G Fronthaul and Midhaul: The Bandwidth Math and What It Means	optics-for-5g-fronthaul-midhaul	tutorial	5G & Telecom	5G fronthaul midhaul eCPRI 25G SR CRAN WDM fronthaul optical latency 50G 100G	5G fronthaul optics eCPRI 25G SR

The optics question in 5G transport gets treated as a straightforward capacity problem — more antenna bandwidth, more fiber, more ports. The reality is more constrained. Fronthaul in particular imposes latency requirements that eliminate certain transceiver types from consideration regardless of their data rate capability, and the bandwidth math for a realistic 5G NR deployment produces numbers that many network planners underestimate until they're deep into a deployment.

The eCPRI Bandwidth Math

The evolved Common Public Radio Interface (eCPRI) specification defines the fronthaul split between a Remote Radio Unit (RRU/RRH) and a Distributed Unit (DU). The bandwidth requirement per sector depends on carrier bandwidth, numerology (subcarrier spacing), MIMO layers, and the compression scheme used.

A 5G NR carrier at 100MHz channel bandwidth with 64 antenna ports (64T64R massive MIMO) using eCPRI Option 7-2x compression requires approximately 25 Gbps of fronthaul capacity per carrier per sector. A three-sector gNodeB with two 100MHz carriers per sector needs 150 Gbps of aggregate fronthaul to the DU. This is why 25G SR is the fronthaul default, not 10G — a single 100MHz 64T64R carrier already exceeds 10G uncompressed, and most deployments use multiple carriers.

The specific math using eCPRI Equation: Required_bps = num_ports × bits_per_sample × sample_rate × IQ_factor × overhead. For 64T64R at 100MHz 5G NR, with 15-bit I and 15-bit Q samples, 30.72 Msps sample rate (3.84 MHz × 8 oversampling), the raw IQ data rate is approximately 59 Gbps. eCPRI Option 7-2x compression targeting 23:1 brings this to around 25 Gbps. With eCPRI overhead and timing messages, 25G links run at around 75% utilization for a single carrier.

At 26 GHz mmWave or mid-band 5G with multiple carriers stacked, this pushes toward 50G and 100G fronthaul requirements even for a single macro site. This is why Nokia and Ericsson have both specified 25G and 100G fronthaul interfaces on their latest generation RRU products.

Why 25G SR Is the Fronthaul Default

The IEEE 802.3by 25GBASE-SR standard specifies multi-mode fiber operation at 850nm with reach up to 70m on OM3 or 100m on OM4/OM5. For fronthaul this means very short links between street-level cabinets or rooftop equipment and the nearby DU equipment. The 25G SFP28 SR module is the standard choice because: the reach is sufficient for most fronthaul topologies, the module cost is substantially lower than 25G LR or ER, and the power consumption (under 1W for a typical SFP28 SR) is manageable in antenna-side equipment with tight power budgets.

The critical constraint for fronthaul optics is not bandwidth — it's latency. The 3GPP specification for 5G NR fronthaul (eCPRI) targets a one-way transport latency of 100 microseconds or less for the HARQ process to work correctly. This 100 µs budget covers all sources of delay: propagation delay on the fiber, serialization delay at 25G, and any switching or processing in the transport network. Propagation delay on fiber is approximately 5 µs/km. A 25G serial link has a serialization delay of roughly 0.04 µs per 125-byte frame — negligible at this link rate.

What this latency constraint rules out is any transceiver type that adds buffering or retiming. WDM-PON and some CWDM aggregation schemes introduce queuing delays that can push the fronthaul latency above the HARQ deadline. For this reason, passive point-to-point fiber or passive WDM (using fixed-wavelength SFP28 modules) is preferred over any active switching layer between RRU and DU.

50G and 100G in Midhaul

Midhaul connects the DU to the Centralized Unit (CU), which handles RRC and PDCP protocol layers. The midhaul bandwidth requirement aggregates multiple DU sites and is therefore higher in total but more tolerant in latency. 3GPP targets are 10 milliseconds for fronthaul-to-midhaul delay, which opens up more transport options.

50G QSFP28 SR (IEEE 802.3cd 50GBASE-SR) has emerged as the midhaul interface for medium-aggregation scenarios: 4 to 8 DU sites converging at a CU. The 50G rate provides headroom for the aggregated fronthaul traffic plus signaling overhead. 100G QSFP28 SR4 or CWDM4 handles larger aggregation nodes where 16 to 32 sectors converge.

For midhaul over longer distances — 10km to 40km between DU aggregation sites and metro CU locations — 25G LR (10km, SMF, 1310nm) and 25G ER (40km, SMF) are widely deployed. The 25G LR SFP28 module draws around 1.5W and is available from compatible vendors at competitive cost. For 100G midhaul over 10km, 100GBASE-LR4 (four-lambda LWDM at 1295-1310nm) is the standard choice.

WDM's Role in CRAN

Centralized RAN (CRAN) architectures that aggregate many RRU sites through passive WDM before reaching the DU pool create specific transceiver selection challenges. Passive CWDM muxes typically support 8 or 18 channels, with channels spaced at 20nm intervals across the O-band and C-band. Each channel uses a fixed-wavelength SFP28 module tuned to its CWDM wavelength.

The CWDM grid for fronthaul is standardized in ITU-T G.694.2. The commonly used fronthaul window spans 1271nm to 1371nm (O-band), supporting 6 channels at 20nm spacing with insertion loss below 1.5 dB per channel for passive mux/demux. This fits 5G NR fronthaul requirements because O-band chromatic dispersion on G.652 SMF is near zero (≈3.5 ps/nm/km at 1310nm), minimizing dispersion penalty at 25G per channel.

A typical CWDM fronthaul installation uses a passive 1×6 or 1×8 CWDM mux at the antenna site, fixed-wavelength 25G SFP28 modules (1271nm, 1291nm, 1311nm, 1331nm, 1351nm, 1371nm) at each RRU interface, and a corresponding demux at the DU aggregation point. Each 25G channel carries one sector's fronthaul traffic. Eight CWDM channels on two fibers (one transmit, one receive) support an 8-sector cell site on a single fiber pair.

The limitation of passive CWDM is fixed channel assignment. If an RRU is moved or reconfigured, the wavelength assignment must be coordinated with the mux port. For dynamic CRAN deployments that expect frequent reconfiguration, tunable DWDM SFP28 modules (typically based on EML or VCSEL designs with thermal tuning) offer wavelength flexibility at higher cost. Tunable 25G DWDM SFP28 modules supporting the full C-band ITU-T 50GHz grid are available from several vendors including ADVA (now Adtran), Lumentum, and compatible suppliers, at roughly 3 to 4 times the price of fixed-wavelength CWDM modules.

Transceiver Selection Checklist for 5G Fronthaul

The practical decision tree for fronthaul optics starts with distance. Under 100m: 25G SR (OM4) or 25G SR (OM3, derated reach). 100m to 500m: 25G BiDi SFP28 (1270/1330nm over single SMF strand, useful where fiber is scarce). 500m to 10km: 25G LR (SMF, 1310nm). Beyond 10km: 25G ER (SMF, 1310nm, class 2 laser safety) or CWDM/DWDM wavelength multiplexed approach.

For all fronthaul applications, avoid any transceiver that introduces buffering or Forward Error Correction (FEC) with latency overhead. The 25G SR and LR families in the SFP28 form factor meet this requirement. Some 25G modules include Reed-Solomon FEC with latencies below 50ns, which is acceptable. Modules advertising "FEC-enhanced sensitivity" with higher latency FEC codes should be validated against the 100 µs fronthaul budget before deployment.

The transceiver question in 5G fronthaul has a clear answer for the dominant deployment scenarios, but the answer changes with scale. A single 100MHz carrier sector uses 25G comfortably. Twenty sectors of 100MHz 64T64R mmWave push the midhaul into 100G territory, and the aggregation point needs 400G. Planning the full capacity cascade before specifying transceivers avoids the upgrade cycle problem.