Introduction

Optical wireless communication (OWC), which employs light waves for data transmission, offers ultra-high bandwidth, rapid data rates, abundant spectral resources, and strong immunity to electromagnetic interference, making it one of the most promising technologies for the sixth-generation (6 G) mobile networks1,2,3,4. The power consumption and carbon emissions of wireless communication networks are projected to increase significantly in the 6 G technology era5,6,7. Therefore, enhancing energy efficiency while maintaining high communication performance has become a critical challenge for the communications industry. A crucial requirement in OWC is the establishment of high-quality wireless optical links between transmitters and receivers, which is closely tied to the energy efficiency of the optical beams. However, conventional beam modulation methods, including mechanically rotating mirrors8 and liquid crystal spatial light modulators (SLM)9,10,11, are inherently limited in beam steering range and often suffer from low efficiency at large angles. This leads to compromised communication quality for users located at large field-of-view (FOV). Optical phased arrays (OPAs) have been extensively investigated for optical beam steering due to their electronic controllability and fast response, particularly with recent advances in integrated photonics12,13,14. Nevertheless, OPA-based beam steering is typically constrained in achieving large-angle deflection primarily along a single axis, and it often suffers from substantial insertion loss, resulting in a relatively low absolute end-to-end efficiency. The recent developed metasurfaces with sub-wavelength scale units, provides a promising solution for achieving large beam deflection angles15,16,17,18,19,20.

Metasurfaces are two-dimensional artificial materials, enabling flexible manipulation of light’s multiple degrees of freedom, including phase, amplitude, polarization, etc21,22,23,24. The sub-wavelength character of metasurface units forms the basis for achieving a large diffraction angle in optical fields25. Nevertheless, regular-shaped metasurface (RSM) structures, such as cylinder or rectangular cuboid, cannot provide the electronic field distributions needed by the high-efficiency large-angle beam deflection. As a result, in a major of metasurface beam deflection works26,27,28,29,30,31,32,33,34, the beam efficiency drops sharply as the steering angle increases, which also severely impact OWC performances at large FOV positions. Although increasing the transmission power can help maintain optical signal reception energy, the exponential energy consumption is an extreme waste of resources. By employing freeform metasurface structures, a significantly expanded parameter space can be explored, which holds great promise for achieving metasurfaces with highly efficient large-angle beam deflection capabilities.

In this work, we propose a two-step shape-optimization method to design the freeform metasurface structures. Different from traditional topology optimization methods35,36,37,38,39,40,41, the first step of our approach involves imposing constraints on the optimization resolution of the freeform structure to ensure the final structure’s minimum feature size within the fabrication capabilities. The optimized metasurface structures achieve beam deflection angles up to 80°, with modulation efficiencies nearly 80% in both forward and reverse directions, highlighting its bidirectional high-performance capabilities. Building on this excellent performance, we developed a metasurface-based high-quality full-duplex video communication system with ultra-low latency, and clearly showcase the remarkable superiority of our optimized metasurface over conventional metasurfaces. Additionally, the SOM-based OWC system is further extended in both transmission distance to 200 m and single-channel speed to 225 Gbps, highlighting its scalability toward long-distance and high-throughput optical communication. Furthermore, we propose an omnidirectional ultra-wide beam deflection scheme based on a fiber array and an optimized structure array, capable of supporting parallel, independent full-duplex communication links for a large number of users.

Results

Metasurface-based OWC systems and the beam deflection efficiency degradation issue

In OWC, if the signal transmission is based on divergent optical beams, as shown in Fig. 1a, it would inherently offer wide coverage without the need for beam steering. This communication approach is well-suited for scenarios where lighting and communication are integrated, such as indoor light fidelity (Li-Fi) systems based on LEDs42,43. However, due to the divergence of light energy, only a small portion of the optical power reaches the receiver, resulting in a significant reduction in communication rates. Moreover, since the propagation distance of the optical field is closely tied to its energy, the communication distance of this approach is also severely limited. In contrast, collimated optical beam links, as shown in Fig. 1b, can establish efficient point-to-point signal transmission, enabling high-speed and long-distance communication (Supplementary Note 1). To achieve wide coverage using this approach, precise and rapid beam steering is required, which imposes strict requirements on the beam modulation techniques. Although several reconfigurable optical frequency metasurfaces have been proposed in recent years44,45, there is still a lack of flexible dynamic modulation approaches.

Fig. 1: Overview of signal transmission methods in metasurface-based OWC system.
Fig. 1: Overview of signal transmission methods in metasurface-based OWC system.The alternative text for this image may have been generated using AI.
Full size image

a Li-Fi systems based on divergent optical beams, offering broad coverage but limited by low power efficiency and short communication range. b Point-to-point OWC using collimated optical beam, enabling high-speed and long-distance transmission with the requirement of precise beam steering for wide-area coverage. c Hybrid RF-optical communication system, where mobile stations connect to the base station via RF links, and base stations interconnect through optical links, leveraging the complementary advantages of RF and optical communication systems. d Wide-angle beam deflection enabled by a quadratic-phase metasurface that collimates Gaussian beams and steers them to different angles. Inset: Top view of the metasurface composed of regular cylindrical unit cells. e Measured deflection efficiency of regular-shaped metasurface beamformers at various angles, demonstrating significant efficiency degradation at large deflection angles. RF radio frequency, BS base station, MS mobile station, FS fixed station.

Beyond all-optical communication, a broader application involves the integration of wireless optical links with existing radio frequency (RF) communication systems46,47,48. As shown in Fig. 1c, mobile stations connect to the base station via RF signals, while the connections between the base stations and the core network are through optical links. This hybrid RF-optical communication system combines the broad coverage capabilities of RF signals with the high bandwidth and high-speed transmission characteristics of OWC, effectively leveraging the complementary strengths of both technologies. In large communication area scenarios such as stadiums and factories, multiple RF base stations can be deployed to cover different areas. These base stations are interconnected via wireless optical links, eliminating the need for complex fiber-optic cabling and associated installation efforts. Whether in point-to-point all-optical wireless communication or in hybrid RF-optical communication systems, maintaining the energy efficiency of each optical link is essential to ensure long communication distance and high communication capacity, especially for receivers positioned within large FOV zones.

To achieve wide-angle collimated beam deflection, a quadratic-phase metasurface can be employed, which not only collimates the divergent Gaussian beams emitted from fibers, but also enables beam deflection to different angles when beam illuminates on different parts of the metasurface32, as shown in Fig. 1d. However, as the deflection angle increases, the deflected beam efficiency would notably decrease, as shown in Fig. 1e. This decrease is primarily due to the difficulty of regular-shape metasurface structures in providing the required transverse momentum for large-angle beam deflection36. As a result, only a portion of the incident energy is converted into the desired deflected beam, while the remainder either passes through the metasurface without deflection or being converted into losses. Therefore, it is necessary to design and fabricate metasurface structures with specialized geometries that can achieve large-angle beam deflection while maintaining high efficiency, a requirement that is difficult to meet with conventional metasurfaces or traditional beam modulators such as lens-based and SLM-based devices. See Supplementary Note 2 for detailed analysis of traditional beam modulation devices limitations.

Two-step shape-optimization method

Topology optimization based on the adjoint method is an effective approach for designing freeform structures with special functionalities, which iteratively converges an initial random refractive index distribution to a final binary structure. The adjoint method allows efficient computation of gradients of a given objective function with respect to multiple design parameters, by using only two simulations, including forward and adjoint simulations. Thus, it is particularly valuable in problems with a large number of design parameters, as it reduces the complexity of computing gradients49. However, conventional topology optimization methods focus solely on maximizing device performance while often overlooking the practical manufacturability of the resulting structures35,36,37,38,39,40,41. As a result, the optimized metasurface structures frequently exhibit features that are difficult to fabricate, as shown in Fig. 2a. These include relatively narrow structures and gaps, small hollow regions and jagged edges. Such fine features may fall below the minimum fabrication tolerance, leading to significant deviations between the fabricated structure and the intended design structure. Consequently, the measured deflection efficiency of the fabricated device may be substantially reduced compared to the simulated performance. More detailed fabrication results and performance loss of conventional freeform metasurfaces are provided in Supplementary Note 3. To address these issues, we propose the two-step shape-optimization method, designed to obtain manufacturable freeform metasurface structures for high-efficiency and large-angle deflections of a 1550 nm beam. This stepwise approach systematically integrates fabrication-compatible structural features into the topology optimization process, while retaining the superior wavefront modulation capabilities of freeform metasurfaces.

Fig. 2: Illustration of the two-step shape-optimization algorithm.
Fig. 2: Illustration of the two-step shape-optimization algorithm.The alternative text for this image may have been generated using AI.
Full size image

a Typical difficulties encountered in the fabrication of the freeform metasurface structures, including narrow gaps, small hollows and jagged edges. b Diagram illustrating the two-step shape-optimization algorithm. The metasurface unit cell has periods Px and Py, with silicon structure height H on a SiO2 substrate. c Curve of the polarization-averaged absolute deflection efficiency variation with the number of iterations, along with visualizations of the structure at specific iteration steps during the optimization process for the 60° SOM beamformer. P: the 400th iteration firstly to apply the suppression mask, Q: the iteration where the absolute deflection efficiency increases again. ADE: absolute deflection efficiency. d Top-view SEM images of the SOM designed for four deflection angles (50°, 60°, 70°, 80°). Inset: Theoretical top-view design structures corresponding to each angle, where the red region represents the Si structure and the light blue region represents the SiO2 substrate.

The proposed optimization algorithm was implemented in MATLAB, with further details provided in the Methods section. Figure 2b illustrates the core concept of the proposed algorithm, which consists of two main optimization stages. In Step I, the process begins with a coarse pixelated design grid, where the pixel size is chosen to align with the minimum fabrication linewidth, thereby inherently avoiding features that are difficult to fabricate. An adjoint-based gradient descent method is employed to calculate the gradient of the objective function and iteratively update the relative permittivity distribution. To enhance the probability of achieving high-performance designs in Step II, a morphological ensemble optimization strategy is subsequently introduced. Specifically, during each iteration, multiple intermediate structures are generated through threshold filtering with different midpoints. The objective function gradients corresponding to these intermediate structures are evaluated and collectively used to guide the update of the permittivity distribution. To reduce the risk of convergence to local optima, a blurring operation is incorporated throughout the optimization process. After several iterations, a low- resolution metasurface structure is obtained, which serves as the initial input for the next step.

In Step II, the design freedom is increased by refining the mesh with smaller pixel size. The fabrication constraints established in Step I are preserved by gradually applying a suppression mask derived from the previous result, modulated through an exponential function during the optimization. To further ensure manufacturability, adaptive morphological operations, including both opening and closing operations, are applied at specific iterations to eliminate fine structural features that are difficult to fabricate. Iterative updates continue until a final shape-optimized metasurface (SOM) is obtained, achieving high-efficiency beam deflection with full fabrication compatibility. This two-step optimization framework, featuring dynamic resolution control and progressive structural suppression, effectively addresses the long-standing trade-off between optical performance and manufacturability, enabling the practical realization of freeform metasurfaces. Additional details of the two-step shape-optimization algorithm are presented in Supplementary Note 4.

Although arbitrary initial permittivity distributions may lead to higher deflection efficiencies after optimization, they often result in an abundance of fine structural features that are difficult to fabricate50. Therefore, as shown in Fig. 2b, we adopt a more regular geometry, a narrow and elongated rectangle, as the initial pattern to guide the optimization process. Additionally, considering the instability of optical beam polarization in practical communication systems, it is essential to ensure high deflection efficiency under arbitrary polarization states. Since any polarization can be decomposed into orthogonal TE and TM components, the two-step shape-optimization method was employed to jointly optimize the deflection efficiency for both TE and TM polarizations, resulting in a polarization-insensitive design. Given that the optimal gradient directions for TE and TM polarizations may diverge, we intentionally remove symmetry constraints in the structure to accommodate asymmetric solutions. In each iteration, a weighted combination of the gradients from both polarizations is computed and used to update the permittivity distribution, enabling balanced improvement across polarization states. The total gradient is computed as:

$${{{{\bf{G}}}}}^{(k)}=\frac{{\eta }_{{{{\rm{TM}}}}}^{(k)}}{{\eta }_{{{{\rm{TM}}}}}^{(k)}+{\eta }_{{{{\rm{TE}}}}}^{(k)}}{{{{\bf{g}}}}}_{{{{\rm{TE}}}}}^{(k)}+\frac{{\eta }_{{{{\rm{TE}}}}}^{(k)}}{{\eta }_{{{{\rm{TM}}}}}^{(k)}+{\eta }_{{{{\rm{TE}}}}}^{(k)}}{{{{\bf{g}}}}}_{{{{\rm{TM}}}}}^{(k)},$$
(1)

where g(k)TE, g(k)TM are the gradients of deflection efficiency with respect to the relative permittivity distribution for TE and TM polarizations, respectively, and η(k)TE, η(k)TM are the corresponding deflection efficiencies at the kth iteration. Taking the optimization of 60° beam deflection efficiency as an example, the low-resolution result (grid pixel size: 50 nm) obtained after the first optimization step is shown in Fig. 2c at the 350th iteration, yielding an average deflection efficiency of 64.96% (TE: 58.32%, TM: 71.60%). Subsequently, the second optimization step is initiated. After a brief increase in efficiency up to point P, a drop is observed due to the activation of the suppression mask at the 400th iteration, which inevitably alters the previously optimized direction. As the optimization continues, the performance improves steadily, and by around 600th iteration, the curve reaches point \(Q\), where the updated structure becomes compatible with the suppression mask. The final design achieves an average deflection efficiency of 81.43% (TE: 72.25%, TM: 90.61%). The deflection efficiencies of the SOMs designed for other angles under TE, TM, and circularly polarized incident light are provided in Supplementary Note 5. During this process, the initially low-resolution structure with sharp edges is simultaneously refined into a high-resolution design (grid pixel size: 10 nm) with smooth contours. Notably, a minimum separation of 160 nm between key structural elements is maintained, significantly above the minimum fabrication linewidth (~50 nm), ensuring the manufacturability. Following the design, the metasurface structures were fabricated, the detailed process of which is provided in the Methods section. Figure 2d presents scanning electron microscopy (SEM) images of metasurfaces designed for deflection angles of 50°, 60°, 70° and 80°, respectively. The corresponding top-view design profiles are shown in the upper right corner. A comparison between the SEM images and the design profiles clearly demonstrates the excellent structural integrity and uniformity of the SOMs, validating the effectiveness of the proposed two-step shape-optimization algorithm in enhancing the fabrication compatibility of freeform metasurface structures. A detailed quantitative analysis of the structural fidelity and fabrication yield of the SOMs is presented in Supplementary Note 3. Additional SEM images for other optimized deflection angles are provided in Supplementary Note 6.

Furthermore, due to the intrinsic symmetry between the forward and adjoint simulations, the optimization process for deflecting normally incident light to large angles shares identical gradient calculations with its inverse process (see Supplementary Note 7 for details). As a result, the optimized structure can not only efficiently deflect vertically normally light to large angles, but also captures light arriving from large angles and redirects it toward the normal direction with high efficiency. This reciprocity forms the theoretical basis for the high-efficiency full-duplex optical communication demonstrated later in this work.

Characterization of shape-optimized metasurface beamformers

The performances of shape-optimized metasurface (SOM) beamformers were comprehensively characterized in terms of optical beam deflection efficiency and communication bit error rate (BER). The experimental setups for measuring forward and backward deflection efficiency are illustrated in Figs. 3a, b respectively. In the forward test, an optical module operating at a wavelength of 1550.12 nm was used as the light source and connected to a single-mode fiber. The output from the fiber was collimated and directed normally onto the SOM, and then modulated and deflected toward the designed angle. In the backward test, the optical path was reversed: a collimated beam illuminated the metasurface at a large incident angle and was modulated into a normally deflected beam.

Fig. 3: Characterization of shape-optimized metasurface beamformers performance.
Fig. 3: Characterization of shape-optimized metasurface beamformers performance.The alternative text for this image may have been generated using AI.
Full size image

a, b Schematic diagrams of the experimental setups for measuring deflection efficiency in the forward (a) and backward (b) configurations. OM: optical module, SMF single mode fiber, SPPS standard photodiode power sensor. c, d Simulated far-field energy distributions obtained through FDTD simulations for forward deflection (c) and backward deflection (d). Results corresponding to SOM and RSM are consistently depicted in red and green, respectively. SOM shape-optimized metasurface, RSM regular-shaped metasurface. e, f Corresponding experimentally observed far-field energy distributions captured using an infrared laser viewing card for forward (e) and backward (f) deflection. g Schematic diagram of the experimental setup for characterizing BER performance and eye diagram measurements. BER bit-error-rate, OM-T optical module transmitter, OM-R optical module receiver, TR transmitter and receiver, BERT bit-error-rate tester, EDFA: erbium-doped fiber amplifier, FOC fiber optical circulator, PC polarization controller. EDT eye diagram tester. h Measured BER-OTP curves for communication links based on SOM and RSM at deflection angles of 60° and 80°, including both forward and backward configurations. Shown below are the receiver eye diagrams measured at an optical transmitted power of 10.5 dBm for the 0° reference link, the 60° SOM forward and backward links, the 80° SOM forward and backward links, and the 60° ROM forward and backward links. OTP optical transmitted power, FW forward, BW backward.

A standard photodiode power sensor (S122Ch, Thorlabs), coupled with an optical power and energy meter (PM400, Thorlabs), was used to measure the optical power. An infrared laser viewing card (VRC2, Thorlabs) was employed to visualize the beam energy distribution. As a representative example, we evaluated the SOM designed for 80° beam deflection, which is the largest deflection angle in our study. Figure 3c, d present the simulated forward and backward beam deflection performances of this metasurface, respectively, using finite-difference time-domain (FDTD) simulations. The red and green curves indicate the normalized optical power distributions as functions of deflection angle for the shape-optimized metasurface (SOM) and the regular-shaped metasurface (RSM), respectively. The beam deflection efficiency is defined as the ratio of the optical power of the deflected beam to that of the incident beam. The experimental results of forward and backward beam deflection are also presented in Fig. 3e, f, respectively. In each figure, the first and second rows show the energy distributions of beams deflected by the SOM and RSM, respectively, across the range of deflection angles. The angular region corresponding to the designed deflection is highlighted by dashed circles. The SOM demonstrates a highly concentrated energy distribution at the designed 80° deflection. In contrast, the RSM exhibits a more dispersed energy profile, with significant intensity distributed toward undesired angle. At the target angle, the SOM achieves an absolute deflection efficiency more than three times greater than that of the RSM, in good agreement with the simulation results. Similarly, in the backward test, the measured deflection efficiency of the SOM closely matches the simulated values and again exceeds that of the RSM by more than three times. These measured deflection efficiencies strongly validate the effectiveness of our two-step shape-optimization strategy to obtain high-efficiency metasurface structures.

Subsequently, the full-duplex communication BER and eye diagram performance of the SOM was evaluated using the setup shown in Fig. 3g. Optical modules (DWDM-SFP25G-10, Cisco) operating at λ1 = 1550.12 nm and λ2 = 1550.92 nm were used to provide the forward and backward working wavelength, respectively. A fiber optical circulator (FSR-1550S-APC, JCOPTIX) was employed to minimize interference between forward and backward communication links, providing isolation values of approximately 54.94 dB from port 2 to port 1 and 64.31 dB from port 1 to port 3. Although the metasurface was optimized for a central wavelength of 1550 nm, simulations revealed that it maintains high beam deflection efficiency across a certain wavelength range around 1550 nm (see Supplementary Note 5 for details). The optical modules were driven by a pseudo-random binary sequence (PRBS) generator integrated on the evaluation board (FPGA KCU116, Xilinx). Each module produced an optical signal carrying a 25 Gbps on-off keying (OOK) PRBS with a sequence length of 231-1. These signals were initially amplified by an erbium-doped fiber amplifier (EDFA1 for the forward link and EDFA2 for the backward link), transmitted through port 1 of a fiber optical circulator, and output via port 2. To mitigate bit error rate (BER) fluctuations caused by polarization instability, polarization controllers were employed. The collimated beams then illuminated the metasurface, either normally in the forward direction after propagating a few millimeters, or obliquely in the backward direction after traveling D = 50 cm through free space. After being deflected by the SOM, the modulated beams were received and coupled into port 2 of a second fiber optical circulator. Finally, the signals passed through a narrowband optical filter, and delivered to the evaluation board for BER measurement and eye diagram test. It should be noted that the BER calculated by the evaluation board is updated in real time, where a value of 10−12 corresponds to an error-free condition (see Supplementary Note 8 for details).

Figure 3h presents the measured bit-error-rate versus optical transmitted power (BER-OTP) curves for communication links at deflection angles of 60° and 80°, along with a 0° reference link. Below the curves, eye diagrams measured at an OTP of 10.5 dBm (corresponding to the black dashed line in the BER-OTP plot) are presented for the 0° reference link, the 60° and 80° SOM forward and backward links, and the 60° ROM forward and backward links. The solid blue curve with circular markers corresponds to the 0° reference, representing the optimal system performance with the lowest BER among all cases. Red and green curves represent the results of SOM and RSM, respectively. Triangular markers denote results at 60°, while asterisks indicate results at 80°. Solid lines correspond to forward deflection, and dashed lines correspond to backward deflection. The results clearly demonstrate that the SOM significantly enhances communication performance, achieving markedly lower BER values and larger eye openings than the RSM under the same optical power. Notably, for backward deflection at 80°, the SOM requires approximately 7 dB less transmitted optical power to achieve the same BER level. Furthermore, the four SOM BER-OTP curves cluster closely around the blue reference curve, and their eye diagrams exhibit similar eye heights and widths. These results indicate that the BER performance of links at large deflection angles closely approach the optimal performance at 0°, thereby confirming their robustness in sustaining high-quality communication. Meanwhile, since the SOM is a passive optical device, no additional time-dependent signal distortion is observed, as detailed in Supplementary Note 9.

High-efficiency, wide-field, full-duplex video communication system

Based on the shape-optimized metasurfaces with excellent inherent capability for bidirectional beam deflection, we constructed a high-efficiency, wide-field, full-duplex video communication system. The communication system adopts a hybrid RF-optical architecture as shown in Fig. 4a, in which a 5 G base station (STS-SDQ001-RU-N41, IPLOOK) is connected to the core network (installed on a server, S423G3, WUZHOU) via a high-efficiency, wide-field optical link, while mobile devices (MD1 and MD2, Mate 30, HUAWEI) access the base station through 5 G signals. The uplink and downlink signal transmission paths are presented by red and orange curves, respectively. This design combines the high-speed and large-capacity advantages of OWC with the high flexibility of RF communication. The corresponding experimental setups of the communication system are shown in Fig. 4b, where the OWC link between the collimator at base station end and the collimator at core network end features a beam deflection angle of 60° and a transmission distance of 2 meters. The OWC link is highlighted in the figure using a red bold line. It is worth noting that beam deflection angles greater than 60° can also be implemented within our communication system. However, as the deflection angle increases, the deflection efficiency of metasurface in the uplink path decreases significantly due to the geometric projection effects (see Supplementary Note 10 for details). Therefore, a deflection angle of 60° with proper energy receiving area was selected for the experiment. This limitation can be readily addressed by increasing the size of the metasurface sample.

Fig. 4: Demonstration of the high-efficiency, wide-field and full-duplex 5 G video communication system.
Fig. 4: Demonstration of the high-efficiency, wide-field and full-duplex 5 G video communication system.The alternative text for this image may have been generated using AI.
Full size image

a Schematic diagram and (b) the actual scene of the large-angle and full-duplex 5 G video communication system. OM optical module, BS base station, MD mobile device, FPGA field programmable gate array, EDFA erbium-doped fiber amplifier, FOC fiber optical circulator, PC polarization controller, RF radio frequency. c Quality comparison of 5 G video call between the communication links based on SOM and RSM respectively. SOM shape-optimized metasurface, RSM regular-shaped metasurface.

During the initial establishment of the video call, MD1 sends a call request to MD2, which requires the transmitting by the base station and scheduling by the core network. Once MD2 receives and accepts the call request, the communication link is established, allowing MD1 and MD2 to transmit video streams captured by their respective camera modules through the same communication link. Specifically, the uplink signal from the mobile device is received by the base station and transmitted through a single-mode fiber to an FPGA for protocol conversion. The signal is then amplified by EDFA1 and emitted from port 1 to port 2 of the fiber optical circulator. The collimated optical signal, which travels over 2 meters in free space, is efficiently deflected by the shape-optimized metasurface and captured by a receiver collimator. After being transmitted from port 3 of the fiber optical circulator to another FPGA for protocol conversion, the signal is sent to the core network. The downlink follows the reverse process. See Supplementary Movie 1 for more detailed information on the working principle of the video communication system.

To demonstrate the critical role of a high-efficiency optical link in communication performance, we conducted a comparative video call experiment in a 5 m × 5 m indoor space, using beam formers based on a shape-optimized metasurface (SOM) and a regular-shaped metasurface (RSM), respectively. To showcase the real-time performance, we used MD1 to capture a looped video and observed the latency and smoothness of the received video stream on MD2. Supplementary Movie 2 shows the entire video call process of the experiment. The SOM and RSM in this experiment are all designed to modulate normally incident beam into 60° deflection direction. As shown in Fig. 4c, under the same transmission power (the EDFA1 maintains a constant output power of 15.7 dBm), the communication link based on SOM has low latency and smooth video calling performance, while the video call based on RSM experiences delays and noticeable stuttering. The major source of the delay primarily originates from degraded optical wireless channel quality, which lowers the received signal-to-noise ratio and triggers TCP retransmissions. The continuous real-time comparative results are presented in Supplementary Movie 3.

Performance enhancement in transmission distance and data rate of SOM-based OWC systems

Building upon the successful indoor full-duplex video-call demonstration, we have extended the SOM-based OWC system toward communication performance metrics of greater practical interest, such as longer transmission distances and higher data rates51,52. In our system, however, long-distance transmission and high-capacity performance are inherently constrained by beam divergence: a small angle divergence enhances power concentration for long-distance transmission, while a large divergence mitigates dispersion-induced angular deviations, ensuring that all channels can be captured by a single receiving lens. Accordingly, we optimized the system separately for long-distance transmission and high-data-rate communication, achieving substantial improvements in each metric.

First, based on the characterization of the long-distance propagation behavior of SOM-deflected beams in Supplementary Note 11, a full-duplex video communication system employing a 200 m transmission-distance, 60°-deflection optical link was established. In addition, Supplementary Note 12 provides an aperture-dependent link analysis of the SOM-based OWC system. Figure 5a shows the schematic diagram of the long transmission distance video communication system. Different from the previously demonstrated indoor video communication system, the long transmission distance system employed separate optical paths to realize full-duplex operation, owing to the limited effective receiving area of the SOM. Meanwhile, to further enhance optical power reception in long-distance communication, alignment modules were implemented at both terminal ends of the system. Further details of the experimental setup and the alignment modules, are provided in and Supplementary Note 13. To assess the practical applicability of the proposed scheme, we carried out a sensitivity analysis of the 200 m, 60°-deflection OWC link, examining the effects of beam divergence, atmospheric attenuation, and deflection stability as detailed in Supplementary Note 14. Figure 5b shows the real scene image of the long transmission distance OWC video communication system during outdoor testing, along with the results of the video call. The panoramic view of the real scene was captured by an unmanned aerial vehicle from above, while video call results were obtained through screenshots from the MD1 and MD2 user interfaces. Before conducting the video call demonstration, we performed BER measurement and eye diagram test of the 200 m, 60°-deflection SOM-based OWC link. Figure 5c presents the measured BER performance and the corresponding eye diagrams at different BER levels. The results demonstrate that an error-free OWC link can be achieved when the optical transmit power exceeds 26 dBm, thereby validating the feasibility of the proposed SOM-based system for long-distance communication. Building on this BER validation, we next demonstrated real-time 5 G video calls over the same 200-m link to further verify system performance under practical application scenarios. See Supplementary Movie 4 for the video call demonstration and an illustration of the working principle of the video communication system. To ensure stable video communication quality, the output power of each EDFA was set to approximately 30 dBm. Owing to the excellent thermal tolerance of the silicon material, under such high optical power illumination, the SOM remains structurally robust without showing any laser-induced damage53,54. See Supplementary Note 15 for a detailed analysis of laser-induced damage risk and thermal drift in shape-optimized metasurfaces for high-power OWC systems. However, because the high optical intensity may pose a potential hazard to human eyes55, appropriate safety precautions were implemented, as detailed in Supplementary Note 16.

Fig. 5: Demonstrations of a full-duplex video communication system enabled by a 200 m, 60°-deflection SOM-based OWC link.
Fig. 5: Demonstrations of a full-duplex video communication system enabled by a 200 m, 60°-deflection SOM-based OWC link.The alternative text for this image may have been generated using AI.
Full size image

a Schematic diagram and (b) actual view of the 200-m 60° deflection full-duplex video communication system. OM optical module, RF radio frequency, BS base station, MD: mobile device, PC polarization controller, EDFA erbium-doped fiber amplifier, FPGA field programmable gate array, BERT bit-error-rate tester, SOM shape-optimized metasurface. c Measured BER-OTP curve together with representative eye diagrams at different BER levels. BER bit-error-rate, OTP optical transmitted power.

Following the demonstration of long-distance transmission, the system’s capacity was enhanced through the implementation of dense wavelength-division multiplexing (DWDM)52. The experimental layout of the SOM-based DWDM system is shown in Fig. 6a, with further details of system design and implementation are provided in Supplementary Note 17. Nine 25 Gbps optical modules (DWDM-SFP25G-10, Cisco) covering channels C29 to C37 (wavelengths from 1547.71 nm to 1554.13 nm, specifically: 1547.71 nm, 1548.51 nm, 1549.31 nm, 1550.11 nm, 1550.89 nm, 1551.70 nm, 1552.51 nm, 1553.32 nm, and 1554.13 nm) were selected as the nine closest to the 1550 nm center wavelength commonly used in DWDM communication, ensuring that each channel maintained high deflection efficiency while minimizing dispersion-induced angular deviations. Each optical module generated a 25 Gbps OOK PRBS signal. The beams were combined by a wavelength-division multiplexer (FMU-D402160M3, FS) and amplified by an erbium-doped fiber amplifier (EDFA1, AEDFA-PA-35-B-FA, Amonics). The resulting composite beam, carrying an aggregate 225 Gbps signal, was emitted from the fiber end facet, collimated, and then deflected to θ = 60° by the SOM. After D = 20 m of free-space propagation, the optical signals were received by a receiver lens and coupled into a fiber. To compensate for power loss in the receiving system, EDFA2 (AEDFA-BO-23-B-FA, Amonics) was employed. The output of EDFA2 was alternatively connected to a spectroscope, used to confirm the multiplexed wavelengths, or to a wavelength-division demultiplexer, which separated the channels for subsequent BER testing. Figure 6b shows the spectrometer scan of the received signal after fiber coupling and amplification by EDFA, where nine distinct wavelength channels from 1547.714 to 1554.126 nm are clearly resolved, demonstrating the stable multiplexing and transmission of all channels. Owing to the unequal output powers of the optical modules at the transmitter, the peak heights of the nine channels exhibit slight variations. Figure 6c presents the measured BER as a function of optical transmitted power (OTP) for each channel. All nine channels demonstrate nearly overlapping BER-OTP curves, indicating that all wavelengths exhibit similar transmission performance. The inset of Fig. 6c shows representative eye diagrams of the channel at 1550.892 nm under different transmitted powers. As the power increases from 17.5 dBm to 22.5 dBm, the eye opening becomes progressively clearer, indicating improved signal quality and reduced noise-induced errors. These results confirm that the proposed SOM-based DWDM system can support nine parallel 25 Gbps channels with reliable transmission over a 20-m free-space link, achieving an aggregate capacity of 225 Gbps in a single 60°-deflection link.

Fig. 6: A 225 Gbps DWDM OWC system enabled by a 20 m, 60°-deflection SOM-based OWC link.
Fig. 6: A 225 Gbps DWDM OWC system enabled by a 20 m, 60°-deflection SOM-based OWC link.The alternative text for this image may have been generated using AI.
Full size image

a Schematic diagram of the SOM-based DWDM system. OM-T optical module transmitter, OM-R optical module receiver, MUX multiplexer, DMUX demultiplexer. b Spectrometer scan results of the optical signal received by the coupling lens at the receiver. NOP normalized optical power. c Measured BER-OTP curves for nine spaced wavelengths ranging from C29 (1547.714 nm) to C37 (1554.126 nm), each of which carries 25 Gbps signals. Inset (from left to right, top to bottom): Receiver eye diagram at the wavelength of 1550.892 nm for optical transmitted powers of (1) 17.5 dBm, (2) 19.5 dBm, (3) 20 dBm and (4) 22.5 dBm.

Design of an omnidirectional ultra-wide FOV OWC system

In the previous video communication systems, a single high-efficiency optical link with large deflection angle were designed for proof-of-concept demonstration. However, the metasurface beamformer can be extended to support multiple wide-angle deflected optical beams by designing a structure array composed of shape-optimized elements for different deflection angles. Figure 7a illustrates the multi-beam deflection component, where a fiber array is integrated with a two-dimensional metasurface structure array consisting of 80 elements, each having a distinct beam deflection modulation angle. Each fiber channel transmits a unique signal, which is modulated by the corresponding shape-optimized structure and deflected towards users in different directions. By carefully designing the deflection angles, the beamformer enables a fully omnidirectional OWC system that supports parallel, independent full-duplex communication links for a large number of users.

Fig. 7: Design of an omnidirectional ultra-wide FOV OWC system.
Fig. 7: Design of an omnidirectional ultra-wide FOV OWC system.The alternative text for this image may have been generated using AI.
Full size image

a Schematic of the multi-beam deflection component, including a fiber array and a two-dimensional metasurface structure array. The four enlarged images on the left show the top-view structures of S1, S2, S3, and S4, with their corresponding deflection angles and azimuthal angles annotated in the form of (θ, φ). The zoom-in image outlined by the red dashed box on the right highlights the detailed layout of θ and φ in the top-right quarter of the array. b Explanation and design principle of θ and φ. The individual metasurfaces S1, S2, S3 and S4 indicated here correspond to those marked in a. c Schematic of the far-field intensity distribution of the outgoing beams from the metasurface array, illustrating different deflection angles as labeled in the figure. d The FDTD simulation result illustrates the energy distribution of the 3D metasurface beamformer projected onto a spherical surface with a radius of 1 meter.

When designing the structure array, we decompose the beam deflection angles of each metasurface structure into polar angle θ and the azimuthal angle φ. Metasurface structures located on square boundaries of the same order relative to the whole structure array center are designed with the same beam deflection polar angle θ. Consequently, we designed four distinct beam deflection polar angles for the 80 structures: 20°, 40°, 60°, and 80°, which are marked in different colors in Fig. 7a. Using the shape optimization method, metasurface structures with these four deflection polar angles and 0° azimuthal angle can be obtained, all exhibiting high beam deflection efficiencies. By rotating these structures in the xy-plane around their respective centers, the azimuthal deflection angle can be adjusted while maintaining the same deflection polar angle. Thus, a full 360° rotation of these structures enables the deflected beams to cover all azimuthal angles. The right side of Fig. 7a labels the beam deflection azimuthal angles designed for structures in the first quadrant of the structure array, which are also the rotation angles needed to obtain these structures from the initial shape-optimized structure. Due to the rotational symmetry of the structure array, the structures in the other three quadrants can be derived by rotating the first quadrant structures by 90°, 180°, and 270° around the whole structure array center, respectively.

To further clarify the design principles of our structure array, four specific structures are selected, labeled as S1, S2, S3, and S4. The schematic diagrams of these structures and their corresponding modulated beam deflection angles are shown in Fig. 7b. First, S1 and S3 have different beam deflection polar angles but share the same azimuthal angle of 0°. Therefore, these two structures can be obtained by performing shape optimization for deflection angles of 60° and 40°, respectively. On the other hand, S1 and S2 share the same deflection polar angle but have different azimuthal angles. S2 can be derived by rotating S1 by 15° around its center on xy-plane. Similarly, S3 and S4 share the same deflection polar angle, and S4 can be obtained by rotating S3 by 22.5° around its center on xy-plane. Through this arrangement, the metasurface array can deflect beams emitted from different fibers in various directions within the half-space, achieving omnidirectional coverage, as shown in Fig. 7c. We conducted FDTD simulations to validate the beam deflection performance of this two-dimensional metasurface array. The simulated far-field beam energy distribution is shown in Fig. 7d, which is consistent with our design.

Theoretically, the number of structures in the two-dimensional array can be increased as needed to achieve denser spatial angular coverage. For instance, a metasurface array consisting of 280 SOMs with deflection angles ranging from 20° to 80° in 10° increments can realize an almost fully covered far-field angular distribution (Supplementary Note 18). This design concept establishes a scalable and reconfigurable framework for omnidirectional optical wireless communication.

Discussion

Although in this work, we have only used the shape-optimization method to optimize the efficiency of a specific large-angle beam, this approach can also be extended to optimize more complex beam deflection modulation structures for different communication functions. As shown in Supplementary Note 19, we optimized a freeform metasurface that modulates vertically incident light to deflect to three distinct angular positions, with the beam deflection efficiency at each angle being adjustable as required. This capability helps reduce the number of optimized structures and is particularly useful in scenarios where energy distribution in the optical link needs to be adjusted based on the number of users connected to each base station. In addition, beyond the Gaussian-beam-based OWC system investigated in this work, the same design strategy can be extended to structured-light communication systems, as discussed in Supplementary Note 20. Beyond its flexible design, the metasurface beamformer also exhibits a significant advantage in compactness, enabling high-performance optical links within millimeter- or even sub-millimeter-scale footprints. This feature is particularly valuable for future highly integrated OWC systems, where the beamformer can be embedded into compact platforms such as small unmanned aerial vehicles (UAVs), facilitating the deployment of agile communication systems or relay stations. Moreover, our design shows the potential for direct integration with light sources such as VCSELs34, which could enable more compact beam deflection device, as discussed in Supplementary Note 21.

To balance the high-speed, high-capacity data transmission performance with the high mobility of the receiver, we employed a metasurface-based RF-optical hybrid communication system, successfully realizing a wireless communication system with properties of high-efficiency, wide-angle coverage, full-duplex, long transmission distance, high transmission rate and multi-channel parallel communication. Although full-optical wireless communication offers higher transmission data rates, the mobility of users would be significantly limited, due to the current technological challenges in flexible dynamic control of optical field with optical-frequency metasurfaces. The small structural periodicity of optical-frequency metasurfaces makes it difficult to incorporate electrodes for pixel-wise phase control like RF metasurfaces56,57,58. Nevertheless, several approaches, such as polarization tuning31, wavelength tuning36, electro-optic modulation in silicon or lithium niobate waveguides array59 and lateral translation tuning60 have been successfully combined with optical-frequency metasurface to achieve dynamic beam control. Each of these tuning methods has its own advantages and limitations. Selecting the most appropriate approach for a given communication scenario enables the SOM to fully exploit its high-efficiency, large-angle beam-deflection capability, as discussed in Supplementary Note 22. Moreover, the dynamic beam-control scheme enabled by the 2D SOM array architecture, based on spatial channel selection through optical switching, provides a complementary and scalable solution for practical communication scenarios, as detailed in Supplementary Note 23.

In conclusion, we propose a two-step shape optimization method, which enables the realization of freeform metasurface beamformer with high beam deflection efficiency (>80%) and large beam deflection angles (~80°), while being easy to fabricate. Based on this metasurface beamformer, we have developed a novel RF-optical hybrid full-duplex wireless communication system, which offers long transmission distance, high communication rates, large communication coverage and superior communication performance, while also ensuring high mobility at the receiver end, and we demonstrated the practical functionality of ultra-low latency high-definition video calls within the system. Supplementary Table S1 presents a comparison between our work and recent representative beam-steering studies, highlighting the superior performance of our approach. Furthermore, by rotating and arranging the optimized structures with different beam deflection angles in a two-dimension layout, we can achieve a multi-user parallel communication system that covers the entire half-space omnidirectionally. The designed metasurface beamformer not only delivers high performance but also boasts a compact form factor, making it easy to integrate into mobile devices such as unmanned aerial vehicles (UAVs). This enables high-mobility OWC relay stations in various complex communication scenarios. Our work offers a brand new technological pathway for the evolution of next-generation wireless communication systems.

Methods

Optimization algorithm and numerical simulation

We implemented the two-step shape-optimization algorithm in MATLAB to optimize the design of the structure for freeform metasurfaces with different functionalities. Given the periodic characteristic of freeform metasurfaces, the rigorous coupled-wave analysis (RCWA) method was used for the numerical simulation of the optical field. Here, we utilized the open-source code RETICOLO as the RCWA solver, which is known for its high speed and accuracy61. All optimizations were performed on a single CPU (Intel® Xeon® Gold 6334 CPU @ 3.60 GHz) with 8 cores and 16 threads. Detailed runtime statistics of the two-step shape-optimization algorithm are provided in Supplementary Note 24. After the design of the SOM structure, we validated their performance using the widely recognized commercial software FDTD Solutions from Lumerical Inc. First, due to the irregular shape of the freeform metasurface structure, we use the import object to create structures quickly and accurately from the cross-sectional diagrams, which can be obtained after optimization in MATLAB. After setting the dimensions and materials of the structure, additional simulation objects are added, including the FDTD simulation region, sources, substrate, and monitors. Periodic boundary conditions were set in the x and y directions, while perfect matching layers are employed along the z-axis to absorb the outgoing waves. The source shape is set as a plane wave, and its type is defined as Bloch/periodic. The material of the substrate is selected as SiO2. The software-integrated object, grating order transmission, was utilized as the monitor to obtain simulation results, including transmission efficiency, outgoing deflection and azimuth angles for each diffraction order. The refractive indices of the fused silica substrate and amorphous silicon were obtained from the Handbook of Optical Constants of Solids (E.D. Palik)62.

Fabrication of the metasurfaces

Before fabrication, a script file is generated to produce the fabrication layout. In this process, MATLAB is employed to write the contours of the shape-optimized metasurface unit cells into line-drawing commands, which include two endpoint coordinates for each line, and are saved in the.scr file. The unit cell structure of the shape-optimized metasurface is first divided into a grid, with the design pixel serving as the fundamental unit. Then edge detection is performed on the grid by evaluating vertical and horizontal edges based on neighboring differences. The code iterates through the grid, comparing adjacent elements to identify discontinuities, which indicate the presence of an edge. For each detected edge, the corresponding endpoint coordinates are computed. Finally, these coordinates are then input into the line-drawing commands. This procedure is repeated for both vertical and horizontal edges, ensuring that edges are fully extracted across the entire grid. The fabrication process began with the deposition of an amorphous silicon film onto a SiO2 substrate using plasma-enhanced chemical vapor deposition. Subsequently, a layer of PMMA A4 resist was spin-coated onto the substrate, followed by the application of a water-soluble conductive polymer (AR-PC 5090). The substrate was then loaded into an electron beam lithography system (Elionix ELS-F125) for pattern exposure. After exposure, the conductive polymer was dissolved in water, and the resist was developed in a solution of MIBK and IPA (1:3), followed by a rinse in isopropyl alcohol. A 40 nm layer of chromium was then deposited via electron beam evaporation to invert the generated pattern, serving as a hard mask for the subsequent dry etching of the silicon layer. The final step involved immersing the samples in a stripping solution of cerium ammonium nitrate to remove the chromium layer, leaving the desired metasurface structure on the substrate.