Based on the increasing number of global users and smart devices, as well as the expectation that the sixth-generation (6G) mobile communication technology is far superior to the fifth generation in terms of peak speed, latency, spectrum efficiency, and so on, the limited frequency band resources in use are facing severe pressure. The feasible solution to this problem is to develop new frequency bands and expand communication bandwidths^[1–3]. As the terahertz (THz) band, from 0.1 to 10 THz, provides an ultrawide frequency range to meet the demand for extremely high data transmission rate within Tbps, THz communication is considered a key technology in 6G^[4–6]. The low-frequency band among the entire THz range, such as the D-band (110 to 170 GHz), is likely to be the first used in 6G radio access networks due to its technological maturity.

Compared to generation by the electric-mixing method^[7–10], the THz signal generated by the photonic-assisted method at the transmitter end has a purer spectrum without harmonic interference and a larger bandwidth^[11]. Using photonic-assisted THz generation technology, some researchers have conducted communication experiments in various frequency bands at ultrahigh rates^[11–15]. However, the bandwidth limitation of electrical components at the receiver end in the full-electronic and photonic-electronic THz communication systems has always been an important factor restricting system performance, and the use of local oscillator (LO) sources at the receiver end is difficult to avoid. The optical-based THz reception technology, which modulates the received THz signal onto the optical carrier and transmits it to a coherent optical receiver, simplifies the hardware of the receiver end as well as integrating it with fiber-optic networks^[16].

In the application scenario of the fiber-THz-fiber communication system based on full photonic up- and down-conversions shown in Fig. 1, in the middle of links where fibers cannot be installed because of unfavorable terrain such as rivers and canyons, or due to earthquakes, volcanic eruptions, and other natural disasters, signals are wirelessly transmitted in the THz frequency band. This system, in which THz transmission and fiber-optic networks complement each other perfectly and seamlessly integrate, is expected to achieve higher transmission rates.

The optical-based THz reception technology employing phase modulators or Mach–Zehnder modulators (MZMs) is far less mature than photonic-assisted THz generation technology. Based on current research progress, Table 1 summarizes the type of modulator used for THz-to-optical conversion at the receiver and achieved results such as net data rate (NDR) of full photonic conversion fiber–THz–fiber communication systems. In the sub-THz frequency band below 100 GHz, a field trial of long-haul E/W-band transmission achieved 26.8-km wireless delivery of a 32-Gbps signal^[17]. Above 200 GHz, the optical-THz-optical link realized up to 200.2-Gbps achievable information rate (AIR) using plasmonic MZMs and Nyquist frequency-division multiplexed (NFDM) technology at around 230 GHz^[21]. At the D-band, the highest NDR currently achieved was only 66.67 Gbps; it employed a newly fabricated thin-film lithium niobate MZM (TFLN-MZM), and the transmission of THz orthogonal frequency-division multiplexing (OFDM) signal over free space was omitted^[19]. There is an urgent need for further exploration and demonstration of high-speed fiber–THz–fiber communication system that utilizes broadband photodiode (PD) and modulator for full photonic conversions at the D-band.

In this paper, a fiber–THz–fiber communication system at the D-band employing full photonic conversion technology was demonstrated, using a lab-based modified unitraveling-carrier photodiode (MUTC-PD) for photomixing at the transmitter and an ultrabroadband commercial packaged TFLN-MZM to modulate the THz signal at the receiver. Over a 10-km standard single-mode fiber (SSMF) at the transmitter, 0.6-m free space, and 5-km SSMF at the receiver, a 33-Gbaud 16 quadrature amplitude modulation (QAM) signal was successfully transmitted, achieving a line data rate of 132 Gbps and an NDR of 116.03 Gbps. In order to approach the system channel capacity, a 31-Gbaud probabilistic shaping (PS) 64-QAM signal was further transmitted, successfully achieving a line data rate of 146.23 Gbps and an NDR of 123.72 Gbps. To our knowledge, this is the highest NDR currently achieved in any full photonic conversion THz system using a commercially packaged TFLN-MZM (see Table 1).

The experimental setup of the fiber–THz–fiber communication system at the D-band based on full photonic up- and down-conversions is shown in Fig. 2. At the transmitter end, the bit data were first mapped into QAM symbols, upsampled, and then passed through a root-raised cosine (RRC) shaping filter before being loaded into a 64-GSa arbitrary waveform generator (AWG). The analog signal output from the AWG drove the IQ modulator and then was modulated onto the light emitted by Laser 1 operating at 193.3985 THz. After being amplified by an erbium-doped fiber amplifier (EDFA 1), the optical signal passed through a polarization controller (PC) to ensure that it could be coupled with the LO light generated by Laser 2 with the highest efficiency through an optical coupler (OC). The coupled light passed through a 10-km SSMF, a variable optical attenuator (VOA), and EDFA 2 to compensate for the loss caused by fiber transmission.

At the optical-to-THz conversion, we used a lab-based D-band photomixer module based on a back-illuminated MUTC-PD, as shown in Fig. 3(a). The frequency responses of the packaged PD under different photocurrents are plotted in Fig. 3(b); they reveal flat output radio frequency (RF) power over the entire D-band. In particular, the photomixer exhibits stable bandwidth performance under different input optical powers. The measured output RF power within the frequency range from 110 to 140 GHz is shown in Fig. 3(c). The saturation photocurrent is 7.5 mA at 120 GHz, which means the module will not saturate until the optical power at point A in Fig. 2 reaches 21.7 dBm, since the responsivity of the photomixer is 0.05 A/W. The optical spectrum at point A is shown in Fig. 2(a). The specific frequency differential between the signal light and the LO light was determined by the signal baud rate. When the signal baud rate was above 30 GHz, the LO light frequency was 193.2755 GHz. The center frequency of the THz signal was, correspondingly, 123 GHz. After photomixing, the D-band THz signal was directly radiated into free space through a horn antenna without electrical amplification. It is worth noting that there are long-distance wireless links in practical applications, and the necessity for high-gain antennas and amplifiers in these systems deserves further research.

After 0.6-m wireless transmission, the THz signal was received by another antenna and then amplified by a low noise amplifier (LNA) and a power amplifier (PA). The amplified received signal was sent to a TFLN-MZM with a 4-dB bandwidth of 110 GHz working at the null point and modulated onto the light emitted by Laser 3 operating at 193.1 THz. The output optical signal was amplified by EDFA 3 and transmitted over a 5-km SSMF. The measured optical spectrum at point B after power control and polarization control is shown in Fig. 2(b). When the signal baud rate was above 30 GHz, the frequency of the receiver’s LO light emitted by Laser 4 was 193.239 GHz. The signal light was coupled with the receiver’s LO light, and a balanced PD was used for coherent detection, downconverting the signal to an intermediate frequency (IF) of 16 GHz. The IF signal was further amplified by an electrical amplifier (EA) and sent to an 80-GSa oscilloscope (OSC). The offline digital signal processing (DSP) sequentially included down-conversion for converting signals from IF to baseband, downsampling, matched filtering, a classic coherent DSP algorithm, demapping, and then calculating the normalized generalized mutual information (NGMI) to evaluate the performance of the communication system. We used the concatenated soft-decision and hard-decision forward error correction coding scheme discussed in Ref. [22], and adopted the corresponding NGMI threshold of 0.921, which was associated with the overall code rate of 0.8790. In the classic coherent DSP algorithm, as the first step, the Gram–Schmidt orthogonal projection (GSOP) was utilized to correct the I/Q imbalance, followed by clock recovery based on the fast square-timing-recovery algorithm. The signal was recovered by equalization based on the cascaded multimodulus algorithm (CMMA). A feedforward carrier phase estimation (CPE) algorithm utilizing a modified quadrature phase shift keying (QPSK) partition approach, was followed by maximum-likelihood detection^[23] to address the phase noise introduced by the independent operation of the lasers. Following CPE, a decision-directed least mean square (DDLMS) algorithm was employed to further enhance the system performance.

The experiment first considered the optimal center frequency of the THz signal for the system. Figure 4(a) displays the variation of NGMI with a center frequency for the 25-Gbaud 16-QAM signal. There are many parameters that are affected by changes in the center frequency, such as the frequency response of the PD, the gain of the amplifier, and the electro-optic response of the TFLN-MZM. It can be seen that the NGMI of the 25-Gbaud 16-QAM signal improves as the center frequency increases from 115 GHz. The optimal center frequency is 121 GHz; then, the performance deteriorates sharply as the center frequency increases further. There are two main reasons for this. First, the intersection of the operating frequency ranges of the PD and amplifiers went from 110 to 150 GHz. Therefore, when the center frequency was too low so that a considerable part of the signal frequency spectrum was below 110 GHz, the imperfect response of the PD and the abnormal operation of the amplifier reduced the signal-to-noise ratio (SNR) and introduced serious intersymbol interference to the system, resulting in severe performance degradation. From the illustration in Fig. 4(a), the received signal spectrum at a center frequency of 115 GHz is shown to be uneven. The increase in the center frequency alleviated this problem. The second reason is the bandwidth limitation of the TFLN-MZM used at the receiving end. When the center frequency was higher, the conversion efficiency from the THz signal to the optical signal achieved through the TFLN-MZM would be lower, directly reducing the optical SNR. It can be observed that there is significant attenuation on one side of the spectrum of the received signal at the 127-GHz center frequency, which is illustrated in Fig. 4(a). At a center frequency of 121 GHz, a balance between frequency responses of the THz device and THz-to-optical conversion efficiency is achieved; the received signal spectrum is also relatively flat. We also tested the optimal center frequency for signals with bandwidth above 30 Gbaud; the frequency was 123 GHz.

Next, we analyzed the influences of optical powers at points A and B on system performance. The effect of optical power at point A on the NGMI of 32-Gbaud 16-QAM signal is shown in Fig. 4(b). Experiments were conducted both under the 0.6-m wireless transmission (back-to-back, B2B) and 10-km SSMF at the transmitter side, and 0.6-m free space and 5-km SSMF at the receiver side transmission (SMF). As the optical power increases, the SNR of the system improves, so the system performance improves. However, this is accompanied by an increase in nonlinear intensity, and the system performance gradually decreases due to the dominance of nonlinearity. The illustrations in Fig. 4(b) show the received constellations with different powers at point A under SMF transmission. With a fixed power of 20 dBm at point A, Fig. 4(c) further shows the variation of NGMI with the change of optical power at point B. Similar to the variation situation and principle with the power at point A, NGMI first increases and then decreases. With the NGMI threshold of 0.921, the system exhibits a dynamic range exceeding 15 dBm.

The experiment tested the maximum baud rate supported by 16-QAM signal transmission to calculate the maximum data rate; the results are shown in Fig. 4(d). The comparison of the 19 and 33-Gbaud signal received constellations illustrated in Fig. 4(d) shows a gradual decrease in SNR. Ultimately, the system achieved 16-QAM signal transmission at a 33 GHz baud rate. The achieved maximum line data rate is 33×4=132Gbps and the NDR is 33×[4−(1−0.8790)×4]=116.028Gbps.

In order to fully utilize bandwidth and power resources and activate more operating points, we further transmitted PS-16-QAM and PS-64-QAM signals and calculated the NDR separately. Figure 5(a) displays the results on baud rate. After applying PS, some important parameters are first rescanned, and then the transmission entropy is adjusted by changing the Maxwell–Boltzmann distribution scaling factor v^[24] at each signal baud rate to make NGMI meet the threshold. Because the system supports the transmission of 19-33 Gbaud 16-QAM signals, only PS-64-QAM signals are transmitted at these baud rates. At the baud rate of 34 GHz, both PS-16-QAM and 64-QAM signals can be transmitted. Figure 5(b) shows the probability distributions of transmitted constellations and received constellations at different baud rates. At a bandwidth of 31 GHz, the PS-64-QAM signal achieves the maximum line data rate and NDR of the system, which are 31×4.7171=146.23Gbps and 31×[4.7171−(1−0.8790)×6]=123.72Gbps, respectively. It should be noted that in practical applications, the additional complexity brought by applying PS technology should be considered.

The NDRs and corresponding scaling factor v values achieved by the 32-Gbaud signals when the optical power at point A changes after adopting PS are shown in Fig. 6(a). Compared to the 16-QAM signal, which could only be transmitted between 19 and 21 dBm of optical power at point A, PS-16-QAM and PS-64-QAM signals can always achieve NDRs of over 70 Gbps when the optical power at point A ranges from 16 to 23 dBm. At low SNR power points, the highest NDRs achieved by PS-16-QAM/64-QAM signals are basically equal, and the PS-16-QAM signal occasionally performs slightly better due to its lower modulation order. At the three power points of 19, 20, and 21 dBm where the 16-QAM signal can be transmitted, the PS-16-QAM signal has the same SNR redundancy as the 16-QAM signal, while the PS-64-QAM signal utilizes the SNR redundancy to achieve a higher NDR than do the 16-QAM signals. As the power at point A increases, the nonlinearity of the system becomes more obvious. The ability to resist the nonlinearity of the PS-64-QAM signal is relatively poor compared to the PS-16-QAM signal, so the achieved NDR will be lower than that of the PS-16-QAM signal. Figure 6(b) shows the transmitted and received constellation probability distributions of PS-16-QAM and PS-64-QAM signals when the optical powers at point A are 16 and 23 dBm, respectively.

A D-band full photonic conversion fiber–THz–fiber communication system was demonstrated. By relying on the D-band high-output-power frequency-response-flat MUTC-PD and broadband commercial TFLN-MZM, we successfully transmitted a 33-Gbaud 16-QAM signal and a 31-Gbaud PS-64-QAM signal through 10-km SSMF, 0.6-m-long wireless distance, and additional 5-km SSMF, achieving a maximum line data rate of 146.23 Gbps and an NDR of 123.72 Gbps. As far as we know, this is the highest record-breaking data rate realized in any millimeter-wave/THz system based on full photonic up- and down-conversion technology that applies commercial TFLN-MZM for THz-to-optical conversion at the receiver end.