On November 13, 2026, NTT announced a "programmable optical circuit" and an accompanying optimization method designed to reduce the digital signal processing (DSP) load on receivers in long-haul optical communications. By reshaping optical signals along the route while they are still light, NTT says it reduced the temporal spread of the received signal from about 27 nanoseconds to about 10 nanoseconds in a transmission experiment over roughly 1,300 km. The research was published in the British scientific journal Nature Photonics on September 25, 2026.

Optical Circuits at Relay Points Shrink Signal Delay Spread Early

The core idea is to have optical circuits shrink signal distortion partway along the route, instead of leaving all of the compensation to the receiver. According to NTT's announcement, the experiment shortened the time range the receiver must compensate for by about 60%. NTT says the number of DSP operations and the amount of memory can be reduced on a similar scale, so it expects the signal processing load to fall by about 60%. That 60% figure is an estimate derived from the reduced signal spread, not a measured reduction in processing.

The prototype circuit has the following specifications:

  • Chip size: 25 mm x 50 mm (fabricated with silica-waveguide PLC technology)
  • Fiber-to-fiber insertion loss: 2.1 dB
  • Operating band: the entire telecom C-band (40 nm), polarization-independent
  • Input/output ports: 8, with fidelity above 96%

Space-Division Multiplexing Drives Up the DSP Burden

The work addresses the growing signal processing demands that come with higher-capacity optical fiber. Space-division multiplexing (SDM), which raises capacity by sending light through a single fiber in multiple propagation patterns (modes), introduces "mode dispersion," in which each mode arrives at a slightly different time. Today this skew is compensated by MIMO-based DSP at the receiver, and according to NTT, the amount of computation is proportional to the square of the number of modes multiplied by the size of the delay difference to be compensated.

With polarization multiplexing, the current mainstream approach, the delay difference to be compensated stays at 1 nanosecond or less. In next-generation long-haul SDM transmission, however, it is expected to reach tens to hundreds of nanoseconds, making DSP power consumption and processing latency a practical barrier. Raising the signal baud rate increases the load in the same way.

Treating the Fiber and Circuits Together as a Neural Network

The optical circuit consists of an array of Mach-Zehnder interferometers, each of which makes two light waves interfere. Heater electrodes on each interferometer adjust the phase, allowing the circuit to freely set how light mixes between its input and output ports.

The optimization approach is also distinctive. NTT treated the entire transmission path, made up of the optical fiber and the optical circuits at each relay point, as a deep neural network, handling each circuit's phase settings like network weights. It then used backpropagation to find settings that minimize the spread of the received signal. NTT says predictions from its computer model of the transmission path agreed quantitatively well with the experimental results.

Verified by Looping Signals Through 51.2 km of Fiber Up to 26 Times

For the experiment, NTT built a recirculating loop setup that repeatedly sends signals through a 51.2 km optical fiber carrying three spatial modes. The transmission distance was varied from 51.2 km (one loop) to 1,331.2 km (26 loops), using a 12 GBd polarization-multiplexed QPSK signal. NTT confirmed similar reductions at multiple wavelengths and says simulations showed the method can also be applied to fibers with more modes.

Our Take: Offloading Work to the Optical Stage Is a Meaningful Idea

This is not a technology that users will notice right away. But as traffic between data centers and over submarine cables grows, receiver power and latency translate directly into operating costs. Against a trend of piling ever more processing onto electronic circuits, this approach of shifting part of the burden to the optical stage offers a different option.

At the same time, these results come from a lab setup using recirculating transmission. In our view, the next question is how well the method can track changes in real deployed cables, such as those caused by temperature. NTT itself lists real-time control that follows changes in transmission path conditions as a possible future direction.

Summary

NTT plans to pursue research and development with a wide range of optical networks in mind, from data center interconnects to links between telecom buildings and intercontinental submarine cables. Beyond compensating for mode dispersion, it is also looking at optimizing transmission characteristics and integrating the circuit with the switching functions of optical nodes. The next steps toward practical use will be worth watching as NTT aims for a future optical network platform that combines higher capacity with lower power consumption and lower latency.