Chinese Team Reports 3.2 Pbit/s Real-Time Transmission over 24-Core Fiber

What the team reported

A Chinese research collaboration has reported aggregate real-time bidirectional transmission of 3.2 petabits per second over a 10.3-kilometre, 24-core optical fiber. The group includes the National Key Laboratory of Optical Communication Technology and Network at China Information and Communication Technologies Group, Pengcheng Laboratory and Fiberhome Fujikura Fiber Optic Technology. The reported system used spatial-division multiplexing together with transmission across the S, C and L wavelength bands. Coverage of the demonstration also says that the S-band portion included 400 Gbit/s-per-wavelength real-time channels.

The figure is an aggregate laboratory capacity, not the speed of one subscriber, one server port or one transceiver. It combines traffic carried in both directions, across many wavelengths and across 24 spatial cores. In practical terms, the work asks how much information can be moved through one cable-sized optical path when engineers expand both the spatial and spectral dimensions. Reports compare the result with an earlier 2.5 Pbit/s demonstration by the team, implying an increase of roughly 28 percent.

The researchers and subsequent media reports describe the result as a real-time transmission record. Public reporting available at publication time, however, does not provide a complete peer-reviewed paper, a full channel plan, error-performance tables or an independent replication. This article therefore treats 3.2 Pbit/s as a reported demonstration result and does not present it as independently verified production performance. That distinction matters because laboratory records normally optimise a specific link, while commercial networks must operate through maintenance events, aging, temperature changes and variable traffic for years.

How 24 cores and three optical bands raise capacity

Conventional single-mode fiber mainly carries signals in one spatial core, and network upgrades usually add more wavelengths or raise the bit rate of each wavelength. Multicore fiber adds another dimension: several optically isolated cores share one cladding. If each core carries a similar set of channels, aggregate capacity can rise without installing 24 separate conventional fibers. The approach may improve cable density in long-haul corridors, metropolitan routes and data-center interconnects where ducts, landing points or conduit space are constrained.

Using the S, C and L bands extends the usable optical spectrum. Commercial dense-wavelength-division systems are commonly centred on the C band and increasingly use the L band; adding the S band creates additional spectral room. That room is only useful when transmitters, receivers, amplifiers, filters and monitoring equipment operate with acceptable noise and gain across the wider range. The reported 400 Gbit/s-per-wavelength real-time S-band operation is therefore a relevant engineering detail: it suggests the experiment was not based entirely on offline processing of captured data.

Capacity does not multiply without penalties. Closely packed cores can couple light into one another, while different cores and wavelength bands may experience unequal loss, dispersion and nonlinear effects. Engineers need fan-in and fan-out devices, multicore connectors, broad-band amplification and digital signal processing that preserve performance across the complete link. They also need a practical way to manufacture, splice and test the fiber consistently. A result over 10.3 kilometres demonstrates the architecture at laboratory scale, but it does not by itself establish the reach, repair procedures or cost of a national transport system.

Commercial relevance and the remaining gaps

The experiment is relevant to optical-module and transport-equipment suppliers because future capacity growth cannot rely indefinitely on faster electronics alone. A commercial multicore, multi-band system would require coherent optical engines, lasers and modulators covering additional bands, integrated multiplexers, low-noise amplifiers, multicore fiber interfaces and test instruments able to observe every spatial and wavelength channel. Packaging and connector suppliers could also gain new work if the industry standardises reliable fan-in, fan-out and field-installable interfaces.

Near-term adoption is uncertain. Existing networks have large installed bases of standard single-mode fiber, C-band line systems, connectors, splicing tools and operational procedures. Multicore fiber must show that its added capacity offsets the cost of new cable, amplifiers, terminals, monitoring, training and spare parts. Operators will also ask whether individual cores can be isolated during failure, how protection switching works, how faults are located and whether repaired links retain the original crosstalk margin.

Power is another constraint. A record assembled from many wavelengths and spatial channels can require substantial lasers, amplifiers and digital processing. Commercial value depends on delivered bits per second per watt, per rack unit and per route kilometre, not only the largest headline throughput. The most credible path may begin in capacity-constrained, point-to-point corridors where both ends can be upgraded together. Broader use will depend on standards, repeatable fiber production, interoperable interfaces and evidence that the system remains stable outside controlled laboratory conditions.

Sources and editorial note

The core facts were cross-checked across TechRadar Pro, Igor’s Lab, Notebookcheck and Tom’s Hardware. These reports consistently identify the 3.2 Pbit/s aggregate rate, 24-core fiber, 10.3 km link and S+C+L operation. Igor’s Lab provides the additional real-time S-band channel detail.

The available sources largely derive from the research team’s announcement and do not replace a complete technical publication. Exact modulation formats, channel counts, forward-error-correction assumptions, test duration, power consumption and performance after field splicing remain important follow-up questions. Comparisons with earlier records can also use different accounting rules, including whether capacity is summed across directions, cores and bands.

Goilips has therefore avoided consumer-download analogies and does not infer product availability from the demonstration. The useful conclusion is narrower: the experiment supplies evidence that spatial and spectral expansion can be operated together in real time at multi-petabit aggregate scale. Turning that evidence into deployable transport will require qualification, standardisation and economic validation across the complete optical system.