The rapid expansion of artificial intelligence (AI) applications has dramatically accelerated the demand for more data centers. These facilities require substantial energy resources for both powering electronic equipment and maintaining cooling systems. To optimize costs, data centers are increasingly located in remote areas away from major urban centers, where construction expenses can be reduced massively and more sustainable energy sources are attainable. However, despite these benefits, latency remains a critical challenge for data center interconnects (DCIs) over long distances^[1]. In conventional standard single-mode fibers (SMFs), light travels through a solid silica-based core, where the propagation speed is constrained by the material’s refractive index. Although efforts have been made to enhance performance using advanced refractive index profiles^[2], these improvements have yielded only marginal gains, leaving latency as a key issue in long-haul data transmission.

Recently, hollow-core fibers (HCFs) have emerged as a breakthrough technology with the potential to effectively address the latency bottleneck. Unlike traditional fibers, HCFs guide light through a central air-filled core, significantly reducing the refractive index of the transmission medium^[3]. This design leads to lower chromatic dispersion and minimized latency^[4,5]. Furthermore, since light interacts minimally with the glass in HCFs, attenuation loss and nonlinearities are greatly reduced^[6–8]. The photonic bandgap fiber (PBGF) was an early type of HCF, but it suffered from relatively high loss^[7]. The nested antiresonant nodeless fiber (NANF) has become the preferred option, owing to its significantly lower transmission loss^[8]. These characteristics result in a reduced need for optical amplification in NANF-based systems to achieve comparable transmission distances. Additionally, in extended band transmission, the suppression of nonlinear effects such as stimulated Raman scattering (SRS) becomes less critical, thereby simplifying SRS management. It is promising that the NANF technology not only ensures high-performance connectivity over vast distances but also aligns with the current trend towards more energy-efficient and sustainably powered data centers. Therefore, NANFs are poised to become a crucial enabler for future low-latency long-haul DCIs. For NANF-based long-haul transmission, the fiber’s length is a key factor that mainly determines the achievable system capacity and distance. Recent research works for NANF-based long-haul transmission with high modulation orders are summarized in Table 1. In Ref. [9], polarization-multiplexed probabilistic constellation shaped 16-quadrature amplitude modulation (PM-PS-16-QAM) wavelength division multiplexing (WDM) with 61 C-band channels at 32 GBaud were transmitted over a 7.72 km NANF, reaching 201 km using the recirculating loop technology. In Ref. [10], 41 C-band channels with PM-PS-16-QAM at 32 GBaud and a 50 GHz channel grid were recirculated for 1150 km with a 11.5 km NANF (three fiber spools spliced together). The attenuation loss coefficient within C-band was approximately 1 dB/km in Ref. [10] and the usage optical spectrum ranged from 1546.12 to 1562.23 nm with a 2 THz bandwidth coverage. However, due to the limited length of fiber span and relatively large attenuation loss, the number of available optical channels and the total capacity were restricted^[9,10]. As a result, the full potential of NANF to support long-haul WDM transmission over a wider bandwidth has not been confirmed yet.

In this paper, we experimentally demonstrate high-capacity and long-haul WDM coherent transmission over a self-fabricated low-loss and long-span 20 km NANF using a recirculating loop technique. The system utilizes 83 optical channels spanning the C-band (1529.163 to 1561.826 nm, 4 THz bandwidth coverage) with a 50 GHz grid and a PM-PS-16QAM signal. The experimental results achieve generalized mutual information (GMI) estimated capacity of 16.763 Tb/s over 1000 km transmission for the first time, to the best of our knowledge. This achievement sets a record for the capacity × distance product in NANF-based long-haul transmission, with a total of 16763 Tb/s × km. Additionally, the maximum achievable transmission distance of the individual optical channels is also investigated, reaching 2000 km for a selected channel.

The experimental setup of the WDM long-haul transmission based on the long-span 20 km NANF recirculating loop is illustrated in Fig. 1. The 83 optical channels are emulated by shaping the amplified spontaneous emission (ASE) noise at a 50 GHz grid using a C-band Finisar Waveshaper. The test channel is modulated and measured by sequentially turning off each of the 83 channels. In the transmitter’s offline digital signal processing (DSP) as shown in Fig. 1(c), digital pseudo-random binary sequence bits are mapped to PM-PS-16QAM with an entropy of 3.5611. After twice up-sampling, the signal is shaped by a square root raised cosine filter with a roll-off factor of 0.01. Then the digital signal is converted to an analog signal by an arbitrary waveform generator (AWG, Keysight M8199 A) operating at 256 GSa/s. The test optical channel light source from a C-band tunable external-cavity laser is modulated by an in-phase quadrature (IQ) modulator, which is driven by the analog signal from the output of the AWG. Subsequently, the test channel is combined with other amplified non-measurement channels via a 10:90 coupler.

Key components of the recirculating loop include two acoustic-optic modulators (AOMs) and a digital time delay generator, which control the transmission timing of the optical signal. The AOMs function as high-speed optical switches. When AOM1 is “ON”, the optical signal passes through AOM1 and one input port of the 2×2 coupler (50:50), entering the fiber recirculating loop. The NANF used in the experiments is 20 km approximately. Figure 1(a) depicts the microstructure of NANF, which contains five nested ring tubes. The outer diameter of the fiber including the cladding is 230 µm. The tube inner diameter is 88 µm and the inner hollow core has a diameter of 30 µm. Both ends of the fiber spans are butt-coupled with SMF pigtails. The connector has a transition loss of about 0.5 dB with negligible back reflection. The attenuation loss across the C-band of NANF is below 0.75 dB/km as illustrated in Fig. 1(b). It exhibits relatively flat loss, with the chromatic dispersion (CD) coefficient ranging from 2.9 to 3.8 ps/nm·km over the wavelength range of 1460 to 1590 nm. After 20 km NANF transmission, a C-band waveshaper is used to maintain the signal spectrum flatness. Then the optical signal is amplified using a C-band optical amplifier (Amonics, AEDFA-C-DWDM-23-B-FA) to compensate for fiber attenuation loss. AOM2 remains “OFF” initially, and when the signal reaches AOM2, AOM1 turns off and AOM2 turns on. It allows the light to continue circulating in the loop. Once the predetermined number of loops is completed, the signal proceeds to the receiver.

At the receiver, an optical tunable bandpass filter (OTBF) is used to filter out the test channel. After going through an amplifier and a polarization controller (PC), the test channel enters the coherent receiver. The coherent receiver contains an optical hybrid, a local oscillator (LO), and four balanced photodiodes. After optical-electrical conversion, a real-time digital oscilloscope (OSC, Keysight UXR0504A) operating at 256 GSa/s captures the polarized signal for further offline DSP. In the receiver’s DSP as shown in Fig. 1(d), the signal is first resampled and synchronized through clock recovery. To mitigate CD accumulated distortions during long-haul transmission, CD compensation in the frequency domain is applied^[11], with the CD coefficient set at 3.5 ps/nm·km. Then, the carrier frequency offset estimation is applied. Since the nonlinearity in NANF is ultra-low^[6], only a 2×2 MIMO (multi-input and multi-output) direct-decision least mean square (MIMO-DDLMS) equalizer is used to correct the residual distortions^[12]. Figure 2 shows the bit error rate (BER) versus different numbers of taps in MIMO-DDLMS for a selected channel. As the number of input taps increases, the BER gradually decreases and reaches a plateau after 51 taps. Therefore, the number of taps is set at 51 to achieve a trade-off between performance and complexity. Afterwards, carrier phase estimation and QAM demodulation are employed to recover the original bit sequences. Finally, the system’s throughput is calculated by the generalized mutual information.

Figure 3 presents the optical spectra of the C-band WDM signal after various transmission distances, including back-to-back (BTB) and distances over 1000, 1200, and 1500 km. These spectra are captured at the receiver using an optical spectrum analyzer (OSA) through a 95:5 coupler. The C-band spans from 1529.163 nm (195.05 THz) to 1561.826 nm (191.95 THz), encompassing 83 optical channels at a 50 GHz grid. As the transmission distance increases, fluctuations in the optical spectra become noticeable. However, these fluctuations are effectively managed by incorporating the WSS within the recirculating loop, helping to maintain signal integrity across the extended distances.

The total capacity of the system estimated by GMI is depicted in Fig. 4(a). The loop is configured for about 1000 km of transmission in the NANF. The baud rate per channel does not exceed 36 GBaud. The BER remains below the soft-decision forward error correction (SD-FEC) threshold of 4×10−2. It is observed that the data rate per channel ranges mostly from 150 to 250 Gb/s. Due to the amplifier limitation, the signal-to-noise ratio (SNR) degrades at both ends of the C-band. The total GMI-estimated capacity for the 83 optical channel WDM transmission based on the NANF is 16.763 Tb/s. It achieves a record of capacity × distance product of 16763 Tb/s × km for NANF-based long-haul transmission. To investigate the maximum achievable transmission distance of individual optical channels under the BER threshold, the signal baud rate is fixed at 36 GBaud. As shown in Fig. 4(b), most optical channels reach distances of up to 1000 km, with one exceeding 2000 km. The varying performance of the optical channels is primarily attributed to the uneven distribution of inter-modal-interference (IMI)^[9] and gas absorption^[13]. The constellations for the optical channels at wavelengths of 1529.948 and 1537.401 nm are shown in the insets of Fig. 4. In inset (i), corresponding to the 1529.948 nm channel, the estimated SNR is at 9.4242 dB, and the achieved data rate is 206.58 Gb/s. The SNR of inset (ii) is 10.679 dB and the data rate is 228.79 Gb/s. It is worth noting that the loss of the NANF used in experiments is approximately four times higher than that of the commercial SMF. To further enhance system capacity, extending the transmission bandwidth into the L-band or even the S-band is a promising direction. However, gas absorption in NANF, particularly from CO2, can introduce significant signal degradation in high-speed coherent transmission over longer distances^[13]. With advancements in the manufacturing process of NANF, component-level improvements, and system-level optimization, we anticipate achieving longer transmission distances and higher capacity in future work.

For transmission latency, it has been measured that the propagation delay in NANFs is 3.354 µs/km, compared to 4.912 µs/km in SMFs^[14]. Over long-haul transmission distances such as 1000 km, this difference translates to a latency reduction of 1.558 ms when using NANFs instead of SMFs. This significant latency saving is crucial for time-sensitive applications and can extend the geographic expansion of data centers.

In summary, we experimentally demonstrate a record-breaking low-latency, high-capacity, and long-haul transmission in a self-fabricated long-span NANF-based optical system. A capacity of 16.763 Tb/s over a 1000 km distance is achieved for the first time by covering the C-band aided with a recirculating loop technique, to the best of our knowledge. With the rapid development of NANF, it is anticipated that NANF-based long-haul optical communication systems will have a transformative impact on the performance and architecture of optical data transport networks in the near future.