Beam manipulation for terahertz communications Authors: Mingxiang Li, Josep M. Jornet, Daniel M. Mittleman, Chong Han Published: 2026 ABSTRACT The terahertz frequency band, ranging from 0.1 to 10 THz, offers extensive spectral resources for next-generation wireless communications systems. To compensate for the limited transmit power of terahertz transceivers and severe propagation losses, high-gain directional transmission is essential, making dynamic beam manipulation a key enabler for practical terahertz communications. The stringent gain requirements further extend the Fresnel region, necessitating efficient beam manipulation across both near-field and far-field conditions. The increasing reliance on beam manipulation reflects a paradigm shift in terahertz communications, where performance scaling is achieved through architecture-level innovation rather than solely through incremental hardware or algorithmic improvements. This article provides a comprehensive overview of terahertz beam manipulation techniques. It begins with an introduction of diffraction theory as the foundational propagation model for beam manipulation. Detailed examples tailored to specific communication scenarios are then presented. Experimental verifications using lenses and metasurfaces are included for three distinct beam manipulation cases. Alternative approaches for achieving beam manipulation, such as reconfigurable intelligent surfaces, are briefly discussed. Introduction Wireless communication systems have become part of our daily life and are indispensable across industries. Future networks are anticipated to support a wide range of demanding applications, from virtual, augmented, and mixed reality to remote control of sensitive operations1–4. The limited spectral resources in the microwave frequencies and the escalating demand for high data capacity, therefore, have prompted the investigation of unexplored frequency resources in higher frequency bands. The terahertz spectrum offers ultra-wide bandwidth and holds immense potential for future 6G and beyond networks. Frequency range covering 0.1–1 THz can theoretically support transmission speeds up to terabits per second level5,6. Recent advancements7–9 further demonstrate the feasibility of terahertz communications in outdoor scenarios with long-distance wireless transmission. These achievements mark a critical step forward for 6G communications, transitioning from laboratory experiments to real-world deployment. However, despite these advantages, terahertz communications face notable challenges, in part due to the lack of high-power solid-state sources10,11, substantial power loss during wireless propagation12,13, and limited availability of energy-efficient transceivers14,15. As the operating frequency increases, terahertz sources suffer from a substantial power decrease as frequencies approaching the terahertz range16. The performance of electronic components such as multipliers and amplifiers degrades considerably at terahertz frequencies, mainly due to fabrication constraints and the parasitic effects17. Conventional CMOS-based devices therefore, generally exhibit suboptimal performance in terms of DC-RF conversion efficiency18,19. Heterogeneously integrated devices can achieve higher power, but their increased fabrication complexity and costs limit mass production viability20,21. For sources above 300 GHz, whether multiplier-chain-based or oscillation-based, the output power of single-source elements typically falls below 1 mW11,22,23. Recent advances, such as a 6 × 6 resonant-tunneling diode array by Canon, Inc., have demonstrated a higher output power over 10 mW at 450 GHz24. However, without power amplification, the output power will rapidly decrease to μW levels at higher frequencies25,26. Photonic generation methods offer an alternative by leveraging optical technologies, typically by photomixing two continuous-wave laser beams at different optical frequencies and down-converting them to terahertz frequencies27,28. While photonic methods can support higher data rates29,30, the output power from photodiodes typically remains lower than that of electronic sources31,32. The limited power of terahertz signals is subject to strong free-space path loss, which increases with the square of both frequency and distance, as described by Friis’ law33. While terahertz transmission also experiences atmospheric attenuation due to interactions with water vapor molecules, multiple low-absorption windows in the terahertz spectrum, such as 220–330 GHz, 625–725 GHz, and 780–910 GHz, have been identified to minimize atmospheric absorption6. Moreover, recent progress also theoretically proven that such molecular absorption can enhance secure wireless links13. Given the limited output power of terahertz sources and high propagation losses, terahertz transmissions are typically highly directional rather than isotropic to counteract these losses. Traditional source-integrated metallic antennas, despite their broader half-power beamwidth34–38, exhibit insufficient gain at terahertz frequencies due to their limited aperture size. Additionally, metal surface roughness induces scattering losses, further degrading the antenna performance39. To enhance the directivity, additional antenna designs such as lens antennas and reflector dishes are often incorporated to expand the aperture37,40,41. Traditional lens antennas and parabolic mirrors can be fabricated through CNC machining or 3-D printing. Although materials like Teflon or other polymers are cost-effective and provide good impedance matching with the sources, their high dielectric losses pose limitations42–44. High-resistivity silicon paired with matching layers offers a promising alternative for terahertz antennas, but challenges in fabrication techniques, such as silicon micromachining and deep reactive ion etching, make it difficult to achieve both compactness and high aperture efficiency45–50. This underscores the need for advanced development of compact, efficient terahertz transceivers. Considering the aforementioned motivations and constraints, terahertz communications systems inherently rely on highly directional transmission between devices. Achieving the high gain required in terahertz transceivers necessitates a larger aperture dimension D, which extends the radiative near-field region. This region is typically known as the Fraunhofer distance, defined as 2D2/λ, where λ is the wavelength. Within this region, wavefronts can exhibit arbitrary shapes, in contrast to the planar wavefronts characteristic of the far-field. Unlike traditional wireless systems that mainly operate in the far-field, recent studies51–54 suggest that practical terahertz networks will likely encompass both near- and far-field conditions, necessitating comprehensive beam manipulation strategies in the novel paradigm of cross near- and far-field regions. For example, for an antenna array with an aperture size of approximately 10 × 10 cm2 operating at 300 GHz, the corresponding near-field distance is about 20 m. As a result, typical indoor scenarios with link distances of 10–20 m fall well within the cross near- and far-field condition. Promisingly coming into this regime, doors are opened to many interesting types of beam manipulation, which can be exploited to enhance channel capacity, mitigate blockage, and support user mobility, as shown in Fig. 1. In the absence of effective beam manipulation, maintaining stable links, enabling flexible beam switching among devices, and supporting multi-user communications would be highly challenging. Therefore, beam manipulation is not merely a performance optimization, but a fundamental requirement for practical terahertz communications. Beyond that, beam manipulation also strongly influences sensing accuracy and efficiency. Naturally, future integrated sensing and communications systems will heavily rely on effective beam manipulation techniques, since controlled beam steering and scanning are essential for both maintaining high-quality communications links and achieving accurate sensing, such as localization and tracking55. This marks a paradigm shift, especially in terahertz communications, in which performance scaling increasingly relies on architecture-level beam manipulation rather than solely on incremental hardware or algorithmic improvements.Fig. 1Beam manipulation for terahertz communications in a comprehensive environment.Beam manipulation for terahertz communications in a comprehensive environment. Diverse beam manipulation strategies enable enhanced channel capacity, blockage mitigation, spatial multiplexing, and support for user mobility. Beam manipulation for terahertz communications in a comprehensive environment. Beam manipulation for terahertz communications in a comprehensive environment. Diverse beam manipulation strategies enable enhanced channel capacity, blockage mitigation, spatial multiplexing, and support for user mobility. Propagation model for beam manipulation To understand the mechanism of beam manipulation, one must first understand how electromagnetic (EM) waves propagate. A common practice is to consider EM waves as vector fields with amplitude and phase, propagating in a homogeneous medium such as air, comparable to light traveling in free space in optics. Here, for simplicity, we temporarily neglect polarization effects. The Huygens-Fresnel principle proposes that every point on a wavefront can be treated as a secondary source with spherical wavelets. These wavelets interfere with one another, and their cumulative effect forms a new wavefront. As a scalar theory, the Huygens-Fresnel principle rigorously calculates the intensity and phase of an electric field in scalar diffraction theory, thus modeling the propagation behavior of EM waves effectively56. Various mathematical models and their approximations and derivations based on the Huygens-Fresnel principle can be found in beamforming or wavefront engineering articles and tutorials56–59. Here, we provide only a summary: assuming a complex electric field at the initial aperture E(x0, y0, 0), the corresponding complex electric field at a propagation distance d can be described by the following expression: 1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E({x}_{1},{y}_{1},d)=\frac{-j}{\lambda }{\sum}_{{x}_{0}}{\sum}_{{y}_{0}}E({x}_{0},{y}_{0},0)\frac{\exp (-jk{r}_{01})}{{r}_{01}}\frac{1+(d/{r}_{01})}{2}\Delta {x}_{0}\Delta {y}_{0},$$\end{document}E(x1,y1,d)=−jλ∑x0∑y0E(x0,y0,0)exp(−jkr01)r011+(d/r01)2Δx0Δy0, where λ is the wavelength, k is the wavenumber, and r01 is the distance between source point (x0, y0, 0) and the observation point (x1, y1, d), expressed as: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${r}_{01}=\sqrt{{({x}_{1}-{x}_{0})}^{2}+{({y}_{1}-{y}_{0})}^{2}+{d}^{2}}$$\end{document}r01=(x1−x0)2+(y1−y0)2+d2. To simplify the analysis, approximations for the near-field and far-field can be employed using Fresnel and Fraunhofer diffraction equations, respectively. By implementing fast Fourier transform, these approximations offer more computationally efficient approaches. Coming into the near-field, where the observation plane is close to the source aperture, Fresnel diffraction provides an expression for the complex electric field in Cartesian coordinates. This can be reformulated in a convolution form as: 2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E({x}_{1},{y}_{1},d)=E({x}_{0},{y}_{0},0)* h({x}_{0},{y}_{0},d),$$\end{document}E(x1,y1,d)=E(x0,y0,0)*h(x0,y0,d), where h(x0, y0, d) is the impulse response of light propagation and can be expressed as: 3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h({x}_{0},{y}_{0},d)=\exp (\,jkd)\cdot \frac{1}{j\lambda d}\cdot \exp (\,j\pi \frac{{{x}_{0}}^{2}+{{y}_{0}}^{2}}{\lambda d}).$$\end{document}h(x0,y0,d)=exp(jkd)⋅1jλd⋅exp(jπx02+y02λd). In the far-field, where the observation distance is sufficiently large, Fraunhofer diffraction simplifies the analysis. Here, a two-dimensional Fourier transform is applied to the complex electric field: 4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E(u,v)={{{\mathcal{F}}}}\{E(x,y,0)\}.$$\end{document}E(u,v)=F{E(x,y,0)}. In this form, the 3-D far-field radiation pattern is represented as an orthographically projected 2-D radiation pattern in terms of u and v, which are expressed as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$u=\cos (\phi )\sin (\theta )$$\end{document}u=cos(ϕ)sin(θ) and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v=\sin (\phi )\sin (\theta )$$\end{document}v=sin(ϕ)sin(θ), respectively. Here, ϕ and θ are the azimuth and elevation angles. Note that u and v can be derived from the spatial-frequency components as u = kx/k and v = ky/k. Using the aforementioned propagation models, we can analyze the behavior of a typical beam with Gaussian amplitude (i.e., beam radius w0 = 10λ) and uniform phase distribution, as depicted in Fig. 2a, b. The analysis explores the beam’s propagation over a distance of 200λ in the near-field through Eqs. (2) and (3)) and examines the resulting far-field patterns through Eq. (4). It is worth noting that the aperture size is expressed in units of wavelength to emphasize the underlying physical behavior rather than a specific operating frequency. The beam radius of w0 = 10λ is chosen as a representative example for clarity and ease of demonstration. Physically, such an aperture size is more representative of terahertz imaging systems. For terahertz communication systems, a larger effective aperture is typically required to generate highly directive beams and compensate for severe propagation loss. Nevertheless, the same propagation characteristics and physical framework discussed here can be directly extended to larger apertures through appropriate scaling. As shown in Fig. 2c, d, as the beam travels in free space, the beam waist expands in the near-field and the initial planar wavefront gradually transforms into a spherical shape. Since the beam starts with a planar phase profile, it exhibits a collimated wavefront, resulting in a high-gain performance. This property leads to a narrow beamwidth in the far-field, as shown in Fig. 2e. The corresponding 1-D far-field patterns at two planes where angles ϕ = 0∘ and ϕ = 90∘ are provided in Fig. 2f.Fig. 2Beam propagation model for a collimated Gaussian beam.a Initial amplitude distribution at the source plane with a Gaussian profile (beam radius w0 = 10λ). b Initial uniform phase distribution at the source plane. c, d Amplitude and phase distributions after a propagation distance of 200λ (approximately one-third of the Fraunhofer distance) in the near-field. e Orthographically projected 2-D radiation pattern. The maximum range of u and v is determined by the resolution, and the range where u2 + v2 > 1 does not have any physical meaning, as they fall outside the beam’s angular extent. f 1-D radiation patterns. Beam propagation model for a collimated Gaussian beam. a Initial amplitude distribution at the source plane with a Gaussian profile (beam radius w0 = 10λ). b Initial uniform phase distribution at the source plane. c, d Amplitude and phase distributions after a propagation distance of 200λ (approximately one-third of the Fraunhofer distance) in the near-field. e Orthographically projected 2-D radiation pattern. The maximum range of u and v is determined by the resolution, and the range where u2 + v2 > 1 does not have any physical meaning, as they fall outside the beam’s angular extent. f 1-D radiation patterns. Concept of beam manipulation In the previous section, we established the propagation model of a given beam, which can be determined by its original complex E-field distribution over a transverse plane. A calculated example of a Gaussian beam with a uniform phase profile propagating in free space was provided. In this section, specific desired propagating beam behaviors are numerically presented through manipulation of the complex E-field distribution. It is worth noting that in conventional microwave and optical systems, beam manipulation techniques often employ both amplitude and phase control, and a wide range of structured beams have been extensively studied, especially in the optical regime60,61. Representative applications range from imaging and microscopy to free-space optical communications62–64. However, in the terahertz range, practical system constraints strongly influence how these concepts can be implemented. Due to the limited power budget in terahertz systems, controlling the amplitude results in substantial power losses65,66, placing a heavy burden on the power budget. Moreover, for a terahertz MIMO systems, phase shifting can be achieved by each element but amplitude control can be achieved only with a large number of elements as a group67,68. As a result, amplitude control is less common in terahertz applications. Instead, phase manipulation is the more widely used approach. The following subsections primarily focus on phase manipulation, assuming a fixed Gaussian amplitude with a beam radius of w0 = 10λ for the source. Although less common, beam control involving both amplitude and phase is briefly discussed, and a comparison is provided. To simplify the analysis, specific operating frequencies are not provided, and the axis will be represented in terms of wavelengths. For clarification, in literature, the y-axis is sometimes defined as the beam propagation direction, since the z-axis is pointing perpendicular to the ground. In this article, we define the transverse plane by x- and y-axes, and the z-axis as the beam propagation direction. The resolution of the Cartesian coordinates will be set at 0.2λ to ensure a sufficiently small secondary point source. Additionally, practical implementations in communication scenarios will be introduced to illustrate how beam manipulation can aid in terahertz communications. Phase manipulation Focused beam In the first case, we consider a short-range high-capacity terahertz communications scenario, such as indoor access links or kiosk-type hotspots, where a receiver requiring maximum power reception for reliable demodulation is located at the center of the transverse plane and operates within the near-field of the transmitter, as shown in Fig. 3a. To maximize the energy at a specific spot in the near-field, a focused beam is necessary. To focus the beam at a point at a distance F from the radiating aperture, the beam’s phase distribution can be employed as follows: 5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi (x,y)={k}_{0}(\sqrt{{F}^{2}+({x}^{2}+{y}^{2})}-F).$$\end{document}ϕ(x,y)=k0(F2+(x2+y2)−F).Fig. 3Analytical results illustrating different beam manipulation strategies.a Focused beam. b Required phase distributions for a focused beam at z = 50λ. c Amplitude distribution at z = 50λ. d Cross-sectional view of the beam propagation. e Far-field radiation patterns. f Multi-focused beam. g Required phase distributions for two-focal-spot beam at z = 50λ. h Amplitude distribution at z = 50λ. i Extended-focal-depth beam. j Required phase distributions for a focal depth from around 50λ–70λ. k Cross-sectional view of the beam propagation. l Flat-top beam. m Required phase distributions for a flat-top beam appear at z = 50λ. n Amplitude distribution at z = 50λ. oBeam shape with letters “SJTU". p Typical phase retrieval algorithm q Required phase distribution for the “SJTU" beam shape. r Amplitude distribution at z = 50λ. Analytical results illustrating different beam manipulation strategies. a Focused beam. b Required phase distributions for a focused beam at z = 50λ. c Amplitude distribution at z = 50λ. d Cross-sectional view of the beam propagation. e Far-field radiation patterns. f Multi-focused beam. g Required phase distributions for two-focal-spot beam at z = 50λ. h Amplitude distribution at z = 50λ. i Extended-focal-depth beam. j Required phase distributions for a focal depth from around 50λ–70λ. k Cross-sectional view of the beam propagation. l Flat-top beam. m Required phase distributions for a flat-top beam appear at z = 50λ. n Amplitude distribution at z = 50λ. oBeam shape with letters “SJTU". p Typical phase retrieval algorithm q Required phase distribution for the “SJTU" beam shape. r Amplitude distribution at z = 50λ. An example of this phase distribution, corresponding to a focal length of 50λ, is shown in Fig. 3b. The resulting beam propagation profile at the focal distance and cross-sectional view are depicted in Fig. 3c, d. As demonstrated, the radiation from the aperture converges to a small focal point. In addition to communications applications, such focused beams are commonly used in imaging and sensing systems69–71, where the small focal spot provides exceptionally high resolution. The smaller diffraction limits enable the detection and resolution of finer details. On the other hand, the focusing performance in the near-field will essentially lead to quicker beam divergence, resulting in a reduced gain performance and an increased beamwidth in the far-field, as shown in Fig. 3e. Multi-focal beam In a following scenario, we consider a multi-user case, such as an indoor access point simultaneously serving two nearby devices. Both devices requiring maximum power reception, positioned symmetrically at a distance from the center and still at distance F from the radiating aperture, as depicted in Fig. 3f. To focus the beam at one focal point offset by ±Δx from the center, the phase distribution can be adjusted as follows: 6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi (x,y)={k}_{0}(\sqrt{{F}^{2}+({(x\mp \Delta x)}^{2}+{y}^{2})}-F).$$\end{document}ϕ(x,y)=k0(F2+((x∓Δx)2+y2)−F). The superposition of two complex electric fields corresponding to separate focal points can help achieve multifocal behavior. Figure 3g, h shows the resulting phase profile and demonstrates that the beam is shaped to target the desired locations. Notably, the superposition modifies the amplitude distribution, but in practice, the source amplitude often retains its original Gaussian distribution. This mismatch introduces imperfections, such as sidelobes, as can be observed in Fig. 3h. The method of simple superposition is an efficient approach for generating beams with a limited number of focal points. For scenarios requiring beams with numerous focal points, more complex methods, such as involving machine learning-assisted algorithms, are recommended72,73. Extended-focal-depth beam In a similar scenario, we consider a single device requires maximum received power but may be located within a range of distances along the propagation direction, for example, due to user movement or position uncertainty in indoor access or kiosk-type applications. The device is positioned at the center but possibly within a range from F1 to F2, as depicted in Fig. 3i. In such a scenario, focusing the beam at a single focal point is insufficient, as it would result in rapid power degradation when the device moves away from the designed focal distance. To ensure robust power delivery and reliable communication across this depth range, an extended focal depth is required. This can be achieved by superimposing the complex electric fields corresponding to focal points at F1 and F2, and adjusting the phase distribution accordingly. An example of the phase distribution designed to provide an extended focal depth from approximately 50λ–70λ is shown in Fig. 3j. As illustrated in Fig. 3k, the focal depth along the z-axis is effectively extended, allowing for consistent power delivery over the specified range. This extended focal depth behavior is also highly advantageous for imaging applications. For instance, conventional raster scanning methods typically provide a 2-D image at a precisely fixed distance and require physical movement of the object for depth exploration71,74. By using a beam with an extended focal depth, the object can be placed within a loose range and potentially, a 3-D image of an object can be captured without needing to reposition the object75–77. Flat-top beam In the next case, we consider a single device equipped with a large receiving aperture, as depicted in Fig. 3l. Such a scenario is representative of near-field high-gain terahertz links employing large-aperture or lens-assisted receivers, where power coupling efficiency, rather than peak field intensity at a single point, becomes the dominant performance metric. To maximize power coupling efficiency78,79, it is ideal to ensure uniform power distribution across the receiver aperture. In other words, a flat-top beam is preferred over the conventional Gaussian beam. To transform a Gaussian beam with a beam radius of w0 into a flat-top beam with a beam radius of w1 at a propagation distance F, a common approach is to employ a phase distribution determined through the geometrical transformation technique80,81. Since the profile is radially symmetric, we provide the 1-D expression in a radial form: 7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi (r)=\frac{2\pi }{\lambda F}{\int }_{\!\!\!\!\! -D/2}^{r}\left[2{w}_{1}\left(\frac{f(r^{\prime} )}{f(D/2)}-\frac{1}{2}\right)-r^{\prime} \right]dr^{\prime} ,$$\end{document}ϕ(r)=2πλF∫−D/2r2w1f(r′)f(D/2)−12−r′dr′,8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f(x)={\int }_{\!\!\!\!\! -D/2}^{x}\exp \; \left(-\frac{2{\zeta }^{2}}{{w}_{0}^{2}}\right)\;d\zeta ,$$\end{document}f(x)=∫−D/2xexp−2ζ2w02dζ, where D is the dimension of the radiating aperture. Here, we provide an example of a conversion to a flat-top beam radius w1 = 15λ at a propagation distance of 50λ. The resultant phase profile is optimized and illustrated in Fig. 3m. Alternative approaches, such as iterative techniques like the Gerchberg-Saxton algorithm, are also available for determining the necessary phase distribution82. This phase profile modifies the Gaussian beam’s intensity distribution, transforming it into a flat-top beam with uniform amplitude over the specified aperture, as shown in Fig. 3n. Such uniform illumination finds extensive applications in super-resolution imaging83 and holography systems84, as it provides spatially consistent excitation and response, thereby improving imaging accuracy compared to non-uniform Gaussian illumination. It is important to note that in the microwave community, a flat-top beam typically refers to a far-field concept characterized by uniform gain performance over a certain angular range85,86. However, in this article, we focus on achieving uniform illumination on a planar screen. Notice that a flat-top beam in the near-field does not necessarily maintain a flat-top profile in the far-field. Arbitrary beam shapes Beyond flat-top beams, Gaussian beams can be transformed into a wide variety of arbitrary spatial field distributions, enabling fine-grained spatial channel shaping for terahertz communications. Such capabilities are particularly relevant to advanced communications scenarios, including spatial multiplexing and interference-aware transmission. This class of beam manipulation techniques is commonly referred to as computational holography77 or holographic MIMO87,88, as illustrated in Fig. 3o. From a classical optical or EM perspective, computational holography provides a general computational framework for synthesizing desired optical or EM field distributions over space, thereby enabling active control of the spatial characteristics of the propagation channel. In principle, holography can be realized through joint amplitude-phase modulation, which offers increased degrees of freedom and improved field-shaping accuracy. However, many holographic implementations in the terahertz band rely on phase-only modulation due to its higher efficiency and relatively simple hardware realization, making it more suitable for practical systems. To generate arbitrary beam shapes, phase retrieval algorithms are typically employed. One widely known approach is the iterative algorithm introduced by Gerchberg and Saxton in 1972, commonly known as the GS algorithm89, which retrieves the phase distribution at a source plane to achieve a desired target intensity profile over another plane. The GS algorithm follows these steps:Initialization: Start with a random phase distribution combined with the source amplitude distribution.Forward propagation: Propagate the wavefront forward to generate a complex wavefront in the target plane.Target amplitude replacement: Replace the resultant amplitude in the target plane with the desired target amplitude distribution while retaining the phase.Backward propagation: Propagate the new wavefront back to the source plane.Source amplitude replacement: Replace the amplitude at the source plane with the initial source amplitude while retaining the updated phase. Initialization: Start with a random phase distribution combined with the source amplitude distribution. Initialization: Start with a random phase distribution combined with the source amplitude distribution. Forward propagation: Propagate the wavefront forward to generate a complex wavefront in the target plane. Forward propagation: Propagate the wavefront forward to generate a complex wavefront in the target plane. Target amplitude replacement: Replace the resultant amplitude in the target plane with the desired target amplitude distribution while retaining the phase. Target amplitude replacement: Replace the resultant amplitude in the target plane with the desired target amplitude distribution while retaining the phase. Backward propagation: Propagate the new wavefront back to the source plane. Backward propagation: Propagate the new wavefront back to the source plane. Source amplitude replacement: Replace the amplitude at the source plane with the initial source amplitude while retaining the updated phase. Source amplitude replacement: Replace the amplitude at the source plane with the initial source amplitude while retaining the updated phase. These steps are iterated until convergence is achieved, at which point the algorithm produces the required phase distribution for generating the target beam shape. A detailed schematic of this process is shown in Fig. 3p. One key limitation of the GS algorithm is its slow convergence after some initial iterations. Due to starting with a random phase distribution, this might lead to convergence to suboptimal phase distributions, potentially resulting in inaccuracies. To address this, various improvements have been proposed, such as the Yang-Gu algorithm90 and the multiple-stage phase retrieval algorithms91,92. These enhancements introduce constraints during the transformation process to improve the convergence of the algorithm and the image quality. It should be clarified that conventional GS algorithm is applied when the target plane is in the far-field, therefore, the propagation model is a simple Fourier transform. In a near-field target plane case, the Fourier transform is replaced with the Fresnel transform to account for near-field propagation effects. Recently, machine learning-assisted phase retrieval algorithms have gained attention for their ability to improve efficiency and accuracy93,94. These methods leverage advanced optimization techniques to overcome the limitations of traditional iterative approaches, offering promising results in beam shaping and related applications95,96. Here, an example of transforming a Gaussian beam into a beam shape containing the letters “SJTU” (Shanghai Jiao Tong University) after a propagation distance of 50λ, is shown below. The phase profile is derived by a modified GS algorithm82, as shown in Fig. 3q, and the resulting field distribution is displayed in Fig. 3r. It should be noted that the algorithm used here is for demonstration purpose and does not represent an optimal implementation. Since the transformation happens in the near-field and certain beam shapes only last for a very short propagation distance, no far-field results will be provided in this case. Nevertheless, such beam-shaping techniques are also applicable in the far-field to generate customized radiation patterns97. Bessel beam In previous cases, we mainly considered line-of-sight communication scenarios. However, in practical terahertz networks, partial blockage by obstacles is common, as illustrated in Fig. 4a. In such cases, conventional highly directive Gaussian beams can suffer severe power loss even when the obstacle blocks only a portion of the beam footprint. Therefore, beam manipulation strategies that enhance link robustness against partial blockage become important. When the obstacle size is relatively small, one potential solution is to utilize a Bessel beam. The amplitude distribution of a Bessel beam can be described by a Bessel function of the first kind, therefore the cross-section of a Bessel beam consists of concentric rings. While an ideal Bessel beam is theoretically infinite and non-diffractive, it is impractical to generate such a beam due to the requirement of infinite energy. Instead, in real-world applications, Bessel beams are non-diffractive only within a finite propagation range. Bessel beams also exhibit a unique self-healing property. Because the beam comprises concentric rings, if a portion of the beam is obstructed by a small object, the pattern will regenerate beyond the obstruction98,99. To generate a Bessel beam with a deflection angle θ, the phase distribution required can be expressed as: 9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi (x,y)={k}_{0}\sqrt{{x}^{2}+{y}^{2}}\sin \theta .$$\end{document}ϕ(x,y)=k0x2+y2sinθ. An example of this phase distribution, corresponding to a deflection angle of 13.8∘, is illustrated in Fig. 4b. The resulting beam profile at a propagation distance of 30λ and the cross-sectional view are shown in Fig. 4c, d. As depicted, a prominent column of constructive interference can be observed along the z-axis, which serves as the primary functional component of the beam. Surrounding this central region are fringing fields that form concentric rings. These rings are an interference pattern created by the superposition of waves. The maximum non-diffraction region is primarily determined by the original aperture size (e.g., w0 for the Gaussian beam) and the deflection angle θ. The maximum distance for the non-diffraction region along the z-axis can be approximated as: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${z}_{\max }={w}_{0}/\tan (\theta )$$\end{document}zmax=w0/tan(θ), which is around 42λ. The non-diffraction region width approximately equals to w0, which is around 10λ along the x- and y-axes. Since the deflection angles are applied symmetrically across the aperture, the far-field pattern shows that the main lobes are located at the deflection angle θ, as illustrated in Fig. 4e. The gain at the broadside direction is therefore reduced from the collimated Gaussian beam case.Fig. 4Analytical results illustrating different beam manipulation strategies.a Bessel Beam. b Required phase distributions for the Bessel beam with a deflection angle of 13.8∘. c Amplitude distribution at z = 30λ. d Cross-sectional view of the beam propagation. e Far-field radiation patterns. f Airy beam. g Required phase distributions for the Airy beam. h Amplitude distribution at z = 80λ. i Cross-sectional view of the beam propagation. Each column is normalized to their own maxima, and the colomap only keeps value >0.95 to indicate the beam propagation trajectory in a simple form. j Far-field radiation patterns. Analytical results illustrating different beam manipulation strategies. a Bessel Beam. b Required phase distributions for the Bessel beam with a deflection angle of 13.8∘. c Amplitude distribution at z = 30λ. d Cross-sectional view of the beam propagation. e Far-field radiation patterns. f Airy beam. g Required phase distributions for the Airy beam. h Amplitude distribution at z = 80λ. i Cross-sectional view of the beam propagation. Each column is normalized to their own maxima, and the colomap only keeps value >0.95 to indicate the beam propagation trajectory in a simple form. j Far-field radiation patterns. Airy beam In the next case, we consider a scenario involving a larger blockage, as illustrated in Fig. 4f. As discussed in the previous section, their non-diffractive and self-healing behavior is confined to a finite propagation range. When the obstacle size becomes relatively large, Bessel beams are no longer able to effectively reconstruct their intensity profile, leading to severe link degradation. In such situations, a potential solution is to employ Airy beams, which exhibit a unique self-accelerating property. As an Airy beam propagates, it follows a curved, approximately parabolic trajectory, allowing the main lobe to bend around obstacles rather than propagate along a straight line. This self-bending behavior enables energy delivery to regions that are geometrically inaccessible to conventional Gaussian or Bessel beams, making Airy beams particularly suitable for non-line-of-sight or heavily obstructed terahertz communications scenarios. The transverse intensity profile of an Airy beam is characterized by a dominant main lobe accompanied by several weaker side lobes distributed along one side of the parabolic trajectory60. In theory, Airy beams can be generated using an amplitude distribution governed by an Airy function100,101. However, such amplitude modulation is difficult to implement in practical terahertz systems. Instead, a widely adopted and implementation-friendly approach is to transform an incident Gaussian beam into an Airy beam by applying a cubic phase distribution102, which can be expressed as: 10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\phi }_{{{\rm{airy}}}}(x,y)=a({x}^{3}+{y}^{3}),$$\end{document}ϕairy(x,y)=a(x3+y3), where a serves as a scaling factor that determines the bending of the trajectory, effectively governing the size of the obstacle the beam can bypass. It is important to note that a larger scaling factor results in a larger curvature but less confined side lobes, which reduces the gain performance in the far-field. To twist the beam back towards the center in the near-field, an additional quadratic phase distribution ϕfocus(x, y) = b(x2 + y2), where b is another scaling factor that determines focal length, can be added to the phase profile. This adjustment effectively rotates the trajectory of the beam to ensure it aligns with the receiver’s position in the targeted position. The final phase distribution can be expressed as the summation of ϕairy(x, y) and ϕfocus(x, y). An example of the final phase distribution, combining linear steering, quadratic focusing, and cubic curving components, is depicted in Fig. 4g. As shown in Fig. 4h, the position of the main lobe shifts during propagation, and after propagation distance of 80λ, an Airy-like intensity distribution emerges. The overall propagation trajectory is illustrated in Fig. 4i, where the beam initially focuses, then curves away from the center, and subsequently returns to the central region after a propagation distance of around 500λ. Notice that each column is normalized to their own maxima, and the colomap only keeps value >0.95 to indicate the beam propagation trajectory in a simple form. The corresponding far-field radiation pattern is also given in Fig. 4j, the main lobe is around the broadside, with a slightly reduced gain compared to normal collimated Gaussian beam. Due to the lack of an accurate mathematical expression for the trajectory, optimizing the curved beam configuration to address specific blockage scenarios remains challenging. Nevertheless, recent progress with machine-learning-assisted strategies demonstrates promising potential for efficiently curving around obstacles103. Amplitude and phase manipulation Anti-focus beam The manipulation of both phase and amplitude also enables unique propagation behaviors. One representative example arises in near-field physical-layer secure terahertz communications scenarios104,105, where two devices are located within the near-field region of the transmitter, as illustrated in Fig. 5a. In this scenario, the closer device acts as a potential eavesdropper, while the farther device is the legitimate receiver. The objective is to suppress signal power at the eavesdropper location while simultaneously maximizing the received power at the intended receiver. This can be achieved using a so-called anti-focused beam, which, as its name implies, behaves oppositely to a focused beam. Instead of concentrating energy in a specific region, the anti-focused beam creates a zone with no power, entirely surrounded by regions of power in all three dimensions. This beam type is also known as a bottle beam106 and is typically generated by the focusing property of a Bessel beam. Considering applying a focusing effect to a Bessel beam that consists of concentric rings, the boundary rings are focused to the first focal point, while the center rings are focused to the second focal point. The detailed principle can be explained by the ray tracing model shown in ref. 107.Fig. 5Analytical results illustrating different beam manipulation strategies.a Anti-focused beam. b, c Amplitude and phase distributions required for the anti-focused beam. d Cross-sectional view of the beam propagation. The transverse plane field distributions at z1--z3 are described by e–g, respectively. The transverse distributions are normalized to their own maxima. h LG beams. i Amplitude and phase distributions for modes when p and l = 0, 1. j OAM beam. k Required phase distributions for a l = 2 OAM beam. l Amplitude distribution at z = 50λ. m Cross-sectional view of the beam propagation. n Far-field radiation patterns for the proposed l = 2 OAM beam. o HG beams. p Amplitude and phase distributions for modes when m and n = 0, 1. Analytical results illustrating different beam manipulation strategies. a Anti-focused beam. b, c Amplitude and phase distributions required for the anti-focused beam. d Cross-sectional view of the beam propagation. The transverse plane field distributions at z1--z3 are described by e–g, respectively. The transverse distributions are normalized to their own maxima. h LG beams. i Amplitude and phase distributions for modes when p and l = 0, 1. j OAM beam. k Required phase distributions for a l = 2 OAM beam. l Amplitude distribution at z = 50λ. m Cross-sectional view of the beam propagation. n Far-field radiation patterns for the proposed l = 2 OAM beam. o HG beams. p Amplitude and phase distributions for modes when m and n = 0, 1. For example, considering a Bessel beam with a deflection angle θ = 13. 8∘, after a propagation of 45λ, the Bessel beam is focused by a phase profile corresponding to a focal length of 20λ. The resultant amplitude and phase distributions are shown in Fig. 5b, c. The beam propagation behavior is illustrated in Fig. 5d, where two distinct focal points are formed along the propagation axis, one in front of the null region and the other behind it, while maintaining a power-free zone in the intermediate region. Such unique propagation behavior can be confirmed by Fig. 5e–g. Such anti-focused beam characteristics offer potential advantages for secure communication by creating a power-free zone that reduces the risk of signal interception, thereby enhancing communications confidentiality. Laguerre-Gaussian beam In the following scenario, we consider a case where the receiver requires maximum channel capacity, as shown in Fig. 5h. A promising method to enhance transmission speed without consuming additional spectral resources is mode multiplexing. These modes are orthogonal to each other, enabling simultaneous transmission of multiple independent data streams over the same frequency band. While previous sections primarily discussed near-field effects, it is important to note that mode orthogonality holds in both near- and far-field regions. These modes can be calculated by solving the Helmholtz equation under paraxial approximation. Under cylindrical coordinates, the complete solutions, incorporating both amplitude and phase, are described by Laguerre-Gaussian (LG) beams. These LG beams at z = 0 are mathematically expressed as108: 11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L{G}_{l,p}(r,\varphi ,0)={C}_{lp}^{LG}{\left(\frac{\sqrt{2}r}{{w}_{0}}\right)}^{| l| }\exp \left(-\frac{{r}^{2}}{{w}_{0}^{2}}\right){L}_{p}^{| l| }\left(\frac{2{r}^{2}}{{w}_{0}^{2}}\right)\exp (jl\varphi ),$$\end{document}LGl,p(r,φ,0)=ClpLG2rw0∣l∣exp−r2w02Lp∣l∣2r2w02exp(jlφ), where l and p represent the azimuthal and radial orders, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{lp}^{LG}$$\end{document}ClpLG is a constant term, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${L}_{p}^{| l| }$$\end{document}Lp∣l∣ is the generalized Laguerre polynomials. A set of LG modes example is illustrated in Fig. 5i. A particularly widely adopted special case of LG beams arises when the radial order p equals zero. In this case, mode orthogonality can be achieved through phase-only manipulation via their helical phase distribution, which is widely known for carrying orbital angular momentum (OAM)109,110, as shown in Fig. 5j. The helical phase distribution can be simplified as: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\phi =\exp (\,jl\varphi )$$\end{document}ϕ=exp(jlφ). An example of the phase distribution for l = 2 is shown in Fig. 5k. The corresponding near-field pattern at a propagation distance of 50λ and the cross-sectional view are shown in Fig. 5l, m. Notably, A distinctive null at the center is observed, which is a characteristic feature caused by their phase singularity. Similarly, in the far-field pattern shown by Fig. 5n, the power null appears at the broadside, indicating that the receiver must either be positioned close to the transmitter or possess a sufficiently large aperture to capture the energy. Therefore, while mode orthogonality remains intact in the far-field, such a transmission system is less likely to function effectively in this region, presenting a practical implementation challenge. In the near-field region, a potential solution to mitigate the beam divergence effect is to apply a non-diffractive OAM beam111. Since OAM can be considered as an additional property introduced to other beams, the phase distribution can be viewed as the superposition of an OAM beam and a Bessel beam, allowing divergence to be suppressed within the non-diffractive region. Hermite-Gaussian beam It is also possible to solve the Helmholtz equation under Cartesian coordinates, and the results can be expressed by Hermite-Gaussian (HG) beams, as shown in Fig. 5o. Hermite-Gaussian modes are a set of orthogonal solutions that describe the transverse field distribution of a Gaussian beam in terms of axial variations, and the Gaussian beam that was earlier discussed is the fundamental mode. HG modes are characterized by two indices: m and n, which represent the number of nodes in the x- and y-directions, respectively. In Cartesian coordinate, these HG beams at z = 0 are mathematically expressed as112: 12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H{G}_{m,n}(x,y,0)={C}_{mn}^{HG}{H}_{m}\left(\frac{\sqrt{2}x}{{w}_{0}}\right){H}_{n}\left(\frac{\sqrt{2}y}{{w}_{0}}\right)\exp \left(-\frac{{x}^{2}+{y}^{2}}{{{w}_{0}}^{2}}\right),$$\end{document}HGm,n(x,y,0)=CmnHGHm2xw0Hn2yw0exp−x2+y2w02, where m and n represent axial orders, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${C}_{mn}^{HG}$$\end{document}CmnHG is a constant term, and Hm and Hn are the Hermite polynomials. A set of HG modes example is illustrated in Fig. 5p. As shown, these modes exhibit symmetry between x- or y-axis, and due to the mutual orthogonality, these modes hold potential in high-speed communications systems as well. Similar to LG mode, power null can be observed and therefore, practical implementation in the far-field region remain challenging. Comparison A comparison of those aforementioned beam types is summarized in Table 1. From a communications-oriented perspective, the advantages of different beam manipulation strategies can be broadly categorized into SNR maximization, robustness improvement, and spatial signaling. Beams designed for SNR maximization generally exhibit low spillover to preserve energy efficiency, whereas beams targeting robustness improvement or spatial signaling intentionally trade energy efficiency for functional benefits, leading to higher spillover levels. For manipulation techniques, generally, phase-only manipulation results in lower complexity, except for some special cases such as arbitrary beam shapes or airy beams, which require rapid phase variation and may increase implementation difficulty. For OAM beams, additional spatial degrees of freedom can be exploited as a potential means to enhance channel capacity. However, achieving different OAM modes within a single phase plate can increase implementation complexity. When both amplitude and phase manipulation are involved, the complexity rises considerably. The table also highlights potential applications in terahertz communications and summarizes the effects in both near- and far-field conditions.Table 1Comparison on different types of beamBeam typeAdvantageApplicationsSpilloverNF and FFManipulationComplexityFocusedSNR MaximizationUser requires maximized energyLowNF: focused FF: reduced gainPhaseLowFlat-topSNR MaximizationMaximized energy coupling efficiencyLowNF: flat-top FF: NAPhaseLowMulti-focusedSNR MaximizationMulti-users for maximized energyLowNF: multi-focused FF: NAPhaseLowExtended focal depthRobustness improvementMaximized energy within a potential rangeMediumNF: focused within a range FF: reduced gainPhaseLowBesselRobustness improvementPassing through small obstacleHighNF: non- diffraction region FF: main lobe at deflection anglePhaseLowAiryRobustness improvementPassing around large obstacleHighNF: curved trajectory FF: reduced gain. Main lobe at around broadsidePhaseMediumAnti-focusedSpatial signalingConfidential communicationsMediumNF: power-free zone surrounded by power in all three dimensions FF: NAAmplitude and phaseHighLaguerre- GaussianSpatial signalingMaximized channel capacityMediumNF: mode orthogonality FF: NA due to power nullAmplitude and phaseHigh(OAM)~ ~ ~ ~ (Phase)(Medium)Hermite- GaussianSpatial signalingMaximized channel capacityMediumNF: mode orthogonality FF: NA due to power nullAmplitude and phaseHighArbitrarySpatial signalingHolographic communicationsHighNF: desired beam shape FF: NAPhaseMediumNF near-field, FF far-field, NA not applicable. Table 1Comparison on different types of beamBeam typeAdvantageApplicationsSpilloverNF and FFManipulationComplexityFocusedSNR MaximizationUser requires maximized energyLowNF: focused FF: reduced gainPhaseLowFlat-topSNR MaximizationMaximized energy coupling efficiencyLowNF: flat-top FF: NAPhaseLowMulti-focusedSNR MaximizationMulti-users for maximized energyLowNF: multi-focused FF: NAPhaseLowExtended focal depthRobustness improvementMaximized energy within a potential rangeMediumNF: focused within a range FF: reduced gainPhaseLowBesselRobustness improvementPassing through small obstacleHighNF: non- diffraction region FF: main lobe at deflection anglePhaseLowAiryRobustness improvementPassing around large obstacleHighNF: curved trajectory FF: reduced gain. Main lobe at around broadsidePhaseMediumAnti-focusedSpatial signalingConfidential communicationsMediumNF: power-free zone surrounded by power in all three dimensions FF: NAAmplitude and phaseHighLaguerre- GaussianSpatial signalingMaximized channel capacityMediumNF: mode orthogonality FF: NA due to power nullAmplitude and phaseHigh(OAM)~ ~ ~ ~ (Phase)(Medium)Hermite- GaussianSpatial signalingMaximized channel capacityMediumNF: mode orthogonality FF: NA due to power nullAmplitude and phaseHighArbitrarySpatial signalingHolographic communicationsHighNF: desired beam shape FF: NAPhaseMediumNF near-field, FF far-field, NA not applicable. Comparison on different types of beam NF near-field, FF far-field, NA not applicable. Experiments of beam manipulation In the previous section, the concept of beam manipulation as well as the theoretical propagation behaviors were introduced. Building on that foundation, this section focuses on the experimental verification of beam manipulation through phase control, a widely used and efficient method in terahertz communications systems. For clarification, this section outlines the general principles and key steps involved in measurement, therefore presents only three typical cases rather than covering all the beam types discussed in the previous section. While various measurement systems are available and many research groups have conducted experiments on beam manipulation46,100,113,114, the aim here is to provide general guidance for such processes rather than showing the state-of-the-art. Methods of implementation In the microwave band, the phased array antenna is the most widely recognized method for beam manipulation. In this approach, power from the source is distributed to an antenna array via a feeding network, with each element incorporating a phase shifter to directly achieve the desired phase distribution. From an optics perspective, a spatial light modulator (SLM) provides a direct method for beam manipulation. SLMs can be adapted for terahertz frequencies by employing optical pumping to modulate the free carrier density in semiconductors, thereby altering their EM properties115,116. However, despite recent advancements, both phased array antennas and SLMs at terahertz frequencies still face practical limitations, primarily due to efficiency challenges117–119. As a result, current practical free-space beam manipulation is predominantly implemented through static methods. A static approach utilizes passive devices to generate a fixed phase distribution tailored to a specific beam manipulation objective. Here, we introduce two typical passive devices for implementation, namely lenses and metasurfaces. Lens-based implementation Lens antennas operate based on traditional optics principles related to the path-length optics theory. As a wave propagates through a dielectric medium, it undergoes a phase delay that depends on the medium’s refractive index nr, the distance traveled l, and the wavenumber k0. This phase delay is given by: 13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\phi }_{delay}=-{k}_{0}{n}_{r}l.$$\end{document}ϕdelay=−k0nrl. Consequently, phase manipulation can be achieved by modifying either the wave’s travel path length or the refractive index of the medium. Conventional lens antennas control the phase by modifying the optical path length through their geometric design40. The height distribution according to the formula: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h=\frac{-\phi }{{k}_{0}({n}_{r}-{n}_{0})}$$\end{document}h=−ϕk0(nr−n0), where n0 is the refractive index of air equals to 1, and nr is the refractive index of the material. Planar lenses are also potential alternatives. These lenses maintain a fixed travel path length but achieve the desired phase variation by leveraging effective medium theory to create an artificial gradient-index. For detailed design principles, readers are referred to refs. 50,120. Material properties and fabrication methods play crucial roles in realizing practical terahertz beam manipulation121,122, as dielectric losses are inherent in most 3-D printed materials. For example, commonly used fused filament fabrication materials such as acrylonitrile butadiene styrene (ABS) and polylactic acid (PLA) exhibit relatively high dielectric losses at terahertz frequencies, with reported absorption coefficients of approximately 10 cm−1 for PLA and 5 cm−1 for ABS around 500 GHz123. As a result, these materials are generally not suitable for large-aperture or thick lens implementations, where cumulative dielectric attenuation becomes non-negligible. In contrast, low-loss polymer materials such as cyclic olefin copolymer (COC) exhibit lower dielectric loss, with an absorption coefficient of only 0.09 cm−1 at around 300 GHz, making them attractive for terahertz applications124. However, processing COC requires higher extrusion temperatures and more stringent fabrication control compared with conventional PLA or ABS, although it remains compatible with standard fused filament fabrication techniques. For commonly used 3-D printing materials such as PLA and COC, the refractive index for both material is approximately 1.5123,124. Metasurface-based implementation Although lenses are the most cost-effective solutions for beam manipulation in the terahertz range, their bulkiness and intrinsic dielectric losses hinder their practical implementation in terahertz communication systems. In contrast, metasurface technology has emerged as a promising alternative for beam manipulation due to its low-profile design and high integrability. Metamaterials, made up from sub-wavelength unit cells, with their ability to effectively manipulate EM waves in terms of amplitude, phase, and polarization125–127. Each unit cell interacts with EM waves behaves like electrical components, such as capacitors, inductors, and resistors. For example, capacitive coupling arises from the separation of conductive patches, inductance is associated with current flow along conductive paths, and resistance accounts for energy loss due to absorption or intrinsic material properties. These components collectively form an equivalent circuit model, explaining the metasurface’s transmission or reflection characteristics, which can be utilized for phase manipulation. Metasurfaces can be fabricated using a variety of materials to achieve diverse functionalities. For instance, dielectric silicon resonators can serve as reflect arrays, enabling quarter-wave or half-wave mirrors128, or as absorbers in specific frequency ranges129. Hybrid designs, such as thin patterned gold films deposited on dielectric substrates, expand the design possibilities for transmit arrays with tailored phase profiles130–133. In metasurface lens designs, precise phase control at the unit-cell level is critical, requiring a phase response that spans at least 2π to construct a discrete Fresnel-equivalent lens. These features make metasurfaces highly promising for terahertz applications, offering enhanced design flexibility and performance enhancements. For more reference related to the concept, design, fabrication and applications of terahertz metasurface, we direct interested readers to the following references121,134–136. Experimental verifications Next, we introduce experimental verifications of beam manipulation reported in the literature, with three typical beam types, including focused beam, Bessel beam, and Airy beam. A typical experimental setup for terahertz beam manipulation using a transmissive device is shown in Fig. 6a. The setup generally employs a vector network analyzer (VNA), equipped with frequency extension modules covering the terahertz band for precise signal measurement. The key modules at the transmitter and receiver sides are described as follows.Transmitter side: A standard gain horn antenna is used to generate a quasi-Gaussian beam. After propagating a certain distance, the beam expands and forms a beam radius of the desired size. At this point, the wavefront is no longer planar due to divergence. To ensure a uniform phase distribution, a collimator is typically introduced.Device under test (DUT): The DUT is positioned adjacent to the collimator and is responsible for introducing the desired phase distribution to achieve the intended beam manipulation. Different DUTs will be discussed in subsequent sections.Receiver side: The observation plane is situated at a distance d from the DUT. The receiver usually consists of a waveguide with a small rectangular aperture. It should be pointed out that the waveguide’s flange coupling may slightly affect its radiation pattern. For ideal performance, a near-field probe with nearly isotropic radiation pattern is recommended to ensure that the received signal is independent of the beam’s incoming direction71,82. The receiver is mounted on a two-axis motorized stage for 2-D scanning of the field distribution. The scanning resolution is set to match the diffraction limit and ensuring precise measurements.Fig. 6Experiments of beam manipulation.a A typical experiment setup. The quasi-Gaussian beam output from the horn antenna experiences a certain propagation distance and its spherical wavefront is transformed into a planar wavefront by a collimator. The DUT performs the beam manipulation, and the beam profile after another propagation distance of d is recorded through a 2-D scanning process. b Focused beam. Refractive index profile and photograph of the fabricated DUT, together with the measured field distributions over 0.2–0.9 THz. Adapted with permission from ref. 137. c Bessel beam measurement. Photograph of the fabricated DUT and its height profile, along with the measured and calculated field distributions. Adapted with permission from ref. 98. d Airy beam measurement. Photograph of the fabricated DUT and the unit cell geometry, together with the measured and calculated field distributions from 0 to 80 mm. Adapted with permission from ref. 102. Transmitter side: A standard gain horn antenna is used to generate a quasi-Gaussian beam. After propagating a certain distance, the beam expands and forms a beam radius of the desired size. At this point, the wavefront is no longer planar due to divergence. To ensure a uniform phase distribution, a collimator is typically introduced. Transmitter side: A standard gain horn antenna is used to generate a quasi-Gaussian beam. After propagating a certain distance, the beam expands and forms a beam radius of the desired size. At this point, the wavefront is no longer planar due to divergence. To ensure a uniform phase distribution, a collimator is typically introduced. Device under test (DUT): The DUT is positioned adjacent to the collimator and is responsible for introducing the desired phase distribution to achieve the intended beam manipulation. Different DUTs will be discussed in subsequent sections. Device under test (DUT): The DUT is positioned adjacent to the collimator and is responsible for introducing the desired phase distribution to achieve the intended beam manipulation. Different DUTs will be discussed in subsequent sections. Receiver side: The observation plane is situated at a distance d from the DUT. The receiver usually consists of a waveguide with a small rectangular aperture. It should be pointed out that the waveguide’s flange coupling may slightly affect its radiation pattern. For ideal performance, a near-field probe with nearly isotropic radiation pattern is recommended to ensure that the received signal is independent of the beam’s incoming direction71,82. The receiver is mounted on a two-axis motorized stage for 2-D scanning of the field distribution. The scanning resolution is set to match the diffraction limit and ensuring precise measurements. Receiver side: The observation plane is situated at a distance d from the DUT. The receiver usually consists of a waveguide with a small rectangular aperture. It should be pointed out that the waveguide’s flange coupling may slightly affect its radiation pattern. For ideal performance, a near-field probe with nearly isotropic radiation pattern is recommended to ensure that the received signal is independent of the beam’s incoming direction71,82. The receiver is mounted on a two-axis motorized stage for 2-D scanning of the field distribution. The scanning resolution is set to match the diffraction limit and ensuring precise measurements. Experiments of beam manipulation. a A typical experiment setup. The quasi-Gaussian beam output from the horn antenna experiences a certain propagation distance and its spherical wavefront is transformed into a planar wavefront by a collimator. The DUT performs the beam manipulation, and the beam profile after another propagation distance of d is recorded through a 2-D scanning process. b Focused beam. Refractive index profile and photograph of the fabricated DUT, together with the measured field distributions over 0.2–0.9 THz. Adapted with permission from ref. 137. c Bessel beam measurement. Photograph of the fabricated DUT and its height profile, along with the measured and calculated field distributions. Adapted with permission from ref. 98. d Airy beam measurement. Photograph of the fabricated DUT and the unit cell geometry, together with the measured and calculated field distributions from 0 to 80 mm. Adapted with permission from ref. 102. Focused beam Building upon the conventional focusing phase profile,137 further extends the design to achieve achromatic focusing by compensating the intrinsic frequency-dependent phase dispersion. Instead of a simple quadratic phase, the required phase profile is jointly engineered across frequencies to ensure that waves at different frequencies accumulate consistent phase delays and converge at the same focal position. To realize this, the metalens is implemented using a spatially varying effective refractive index distribution. As shown in Fig. 6b, the continuous index profile is discretized into concentric regions and mapped to subwavelength dielectric unit cells with different geometrical parameters, enabling precise phase control over a broad frequency range. Experimental results shown in Fig. 6b demonstrate that a stable focal spot is maintained over a wide frequency range (0.2–0.9 THz), with the measured field distributions showing consistent focal positions and tightly confined beams. Bessel beam Building upon the Bessel beam phase profile, the required phase distribution is implemented using an axicon-type structure.98 realizes such a phase profile by a dielectric axicon, which transforms an incident collimated beam into a non-diffractive Bessel beam through conical wavefront modulation. As shown in Fig. 6c, the measured field distributions at 300 GHz demonstrate a clear non-diffractive region with a nearly invariant beam width over an extended distance. The transverse intensity profile further confirms the formation of a central main lobe surrounded by concentric sidelobes, which is a typical feature of Bessel beams. Airy beam Building upon the Airy beam phase profile, a low-profile metasurface is designed to generate an Airy beam at 400 GHz.102 reports a folded topology consisting of a single-layer reflective metasurface and a polarizing grid to realize the required phase modulation for Airy beam, as shown in Fig. 6d. The measured field patterns observed at multiple planes up to 80 mm from the metasurface are also presented. As expected, the generated beam exhibits a pronounced main lobe accompanied by multiple sidelobes along both transverse directions. With increasing propagation distance, the main lobe follows a parabolic-like curved trajectory, initially shifting slightly downward and then bending upward, while the sidelobe structure becomes more prominent. This behavior is consistent with the characteristic propagation of Airy beams. Communications demonstrations Finally, we investigate beam manipulation in practical communications scenarios through representative experimental demonstrations. In particular, two typical beam types, including the Bessel beam and the Airy beam are considered, highlighting their unique capabilities in mitigating blockage and enabling non-line-of-sight transmission. Specifically, Bessel beams are more suitable for scenarios with small or partial obstructions located near the beam center, where their self-healing property allows the beam to reconstruct after blockage. In contrast, Airy beams are advantageous in scenarios where the transmitting aperture is fully blocked on one side, as their self-accelerating and curved propagation enables the signal to bypass obstacles. Bessel beam To demonstrate the self-healing capability of the Bessel beam, communications experiments are conducted by introducing an obstacle along the propagation path. As reported in ref. 99, a Bessel-beam terahertz link operating at a 130 GHz carrier frequency with signal bandwidths of up to 20 GHz is experimentally evaluated. The Bessel beam is generated using a 3D-printed axicon. The experiment is performed over a 40 cm transmission distance, with an absorbing obstacle placed 20 cm away from the transmitter, directly blocking the central main lobe region of the beam. The obstacle consists of a hollow plastic cylinder with a 5 cm diameter and 2 cm height, filled with water-soaked fabric. As shown in Fig. 7a, under blockage conditions, the constellation diagram corresponding to the Bessel beam remains well-defined, whereas the Gaussian beam exhibits a severely distorted constellation. This observation is consistent with the reported SNR degradation, where the Bessel beam experiences only a few dB loss, while the Gaussian beam suffers a reduction of up to 14 dB. Moreover, for 20 GHz bandwidth signals, the Gaussian-beam link becomes undetectable, while the Bessel-beam link remains detectable. These results indicate that Bessel beams can effectively mitigate blockage effects in terahertz links, making them a promising candidate for robust communication in dynamic or obstructed environments.Fig. 7Communications demonstrations of beam manipulation.a Self-healing property of the Bessel beam. A small obstacle is placed between the axicon and the receiver, and the measured constellation diagrams of the Bessel beam are compared with those of a Gaussian beam. Adapted with permission from ref. 99. b Curved-beam property of the Airy beam. A metallic obstacle is placed between the phase plate and the receiver, and the measured eye diagrams of the Airy beam are compared with those of a Gaussian beam. The received power, measured BER, and power degradation are also presented. Adapted with permission from ref. 100. Communications demonstrations of beam manipulation. a Self-healing property of the Bessel beam. A small obstacle is placed between the axicon and the receiver, and the measured constellation diagrams of the Bessel beam are compared with those of a Gaussian beam. Adapted with permission from ref. 99. b Curved-beam property of the Airy beam. A metallic obstacle is placed between the phase plate and the receiver, and the measured eye diagrams of the Airy beam are compared with those of a Gaussian beam. The received power, measured BER, and power degradation are also presented. Adapted with permission from ref. 100. Airy beam To demonstrate the self-accelerating property of the Airy beam, communications experiments are conducted by introducing a large obstacle along the propagation path. As reported in ref. 100, a curved-beam terahertz link is experimentally demonstrated at a carrier frequency of around 200 GHz. The communications experiment is performed using OOK modulation at 310 Mbit/s. The Airy beam is generated using a transmission-type phase plate, and the receiver is placed 1 m away from the transmitter with a lateral offset of 10 cm. In the experiment shown in Fig. 7b, without any phase modulation, no signal is received at the off-axis receiver location. When a linear phase plate is used to steer the beam toward the receiver, a received power of −18.4 dBm with a BER of 1.8 × 10−7 is achieved, and a clear eye diagram can be observed. When a metallic obstacle is introduced into the propagation path, blocking the transmitting aperture on one side, the received power drops by about 5 dB, leading to a pronounced degradation in BER, and the eye diagram becomes severely closed. By replacing the linear phase plate with a caustic phase plate that enables a curved propagation path bypassing the obstacle, approximately 3.6 dB of power is recovered, and the BER is reduced by three orders of magnitude, while the eye diagram becomes clear again. These results demonstrate that self-accelerating terahertz beams can effectively bypass obstacles and maintain communication links even when the direct line-of-sight path is blocked. Programmable metasurface towards reconfigurability Reconfigurable intelligent surface Still building on the metasurface principle but one step forward, active semiconductor components or tunable materials can be integrated into the unit cell structure, leading to the design of reconfigurable intelligent surfaces (RIS), also known as programmable metasurfaces. By enabling programmable control of the EM environment, RIS has been explored not only for improving signal coverage and communications performance, but also for emerging applications such as integrated multifunctional platforms including ISAC and wireless power transfer, as well as intelligent metasurface systems with sensing, storage, and computation capabilities. Since these active components require power inputs and wiring connections, they are typically implemented in a reflective mode rather than transmissive. In this article, we briefly review two common types of RIS, namely, semiconductor-based and tunable material-based. For more detailed discussions on these types of RIS, or other types based on micro-electromechanical system (MEMS), mechanical tuning and space-time modulation, we recommend consulting the referenced materials138–142. On one hand, semiconductor-based metasurfaces utilize active components such as diodes and transistors. In principle, applying a bias voltage to diodes, such as PIN and varactor diodes, adjusts the resistance or capacitance of the metasurface elements, therefore achieving a different phase response and enabling real-time phase modulation143,144. Transistors, such as high-electron-mobility transistors, which operate more efficiently at terahertz frequencies, enable better stability and high-speed modulation, making them suitable for terahertz communications systems145–147. In addition, semiconductor-based metasurfaces can be developed into on-chip configurations, offering more compact, highly integrated solutions for ultrafast modulation in terahertz systems. These on-chip metasurfaces integrated with tunable components are also suitable for applications in sensing and imaging applications. Representative demonstrations148 include a 300 GHz tiled CMOS chip with 8-bit phase control, enabling dynamic beamforming over a ±30∘ range, which operates at 1.2 V with a reported static power consumption of 240 μW, as well as a 98 × 98 unit CMOS array achieving a 1∘ beamwidth at 265 GHz, with a static power consumption of 200 μW and a total power consumption of around 850 μW149,150. Recent progress also suggests that semiconductor-based metasurfaces can further enable opto-electronic integration. Representative architectures combine photodiodes and Schottky diodes to support bidirectional optical-EM signal mapping. Although current demonstrations are mainly reported at microwave frequencies151,152, this approach shows promising potential toward terahertz systems, enabling cross-media wireless communications in diverse environments. On the other hand, tunable material-based metasurfaces leverage advanced materials like liquid crystals (LC) and vanadium dioxide (VO2) to modulate EM properties. Liquid crystals, with their birefringence properties under varying bias voltages, i.e., the equivalent dielectric constant of the LC, and therefore the phase response, can be changed according to the voltage applied. Such tunable behavior enables precise phase control in applications such as high-resolution beam scanning, with a low cost and well-established fabrication processes66,153. Recent progress154 demonstrated dual-broadband LC-based programmable metasurfaces enabling reconfigurable beam steering in terahertz wireless communication systems. VO2, due to its unique material characteristics, is capable of switching between insulating and metallic states through electrical, optical, or thermal stimulation. The insulator-metal transition property also provides versatile and cost-effective solution for dynamically tunable phase distribution155,156. Comparison While RIS technologies, as well as the lens and metasurface designs, have all shown effective beam manipulation capabilities, it is important to recognize that each method has its own advantages and limitations. The selection of an appropriate approach depends on specific conditions and requirements. To facilitate this, we provide a general comparison of these three technologies in Table 2.Table 2Comparison on different beam manipulation strategiesTypeLensMetasurfaceRISPhase quantizationContinuousDiscretized (at least 3-bit)Discretized (typically 1-bit)BandwidthLarge (Achromatic level)Moderate (Limited due to dispersion)Small (Oscillation level)EfficiencyModerate (Loss depends on bulkiness)High (Tailored matching impedance)Low (Loss due to less efficient active components)Aperture sizeModerate (Limited by bulkiness)Large (No strict limitation)Small (Limited by manufacturing)CompactnessLowModerateHighReconfigurabilityStaticStaticDynamicCostLowModerateHighNo specific numerical values are presented here, since the types and characteristics of the strategies can vary depending on the context and conditions. This comparison aims to highlight the general trends rather than provide precise data. Table 2Comparison on different beam manipulation strategiesTypeLensMetasurfaceRISPhase quantizationContinuousDiscretized (at least 3-bit)Discretized (typically 1-bit)BandwidthLarge (Achromatic level)Moderate (Limited due to dispersion)Small (Oscillation level)EfficiencyModerate (Loss depends on bulkiness)High (Tailored matching impedance)Low (Loss due to less efficient active components)Aperture sizeModerate (Limited by bulkiness)Large (No strict limitation)Small (Limited by manufacturing)CompactnessLowModerateHighReconfigurabilityStaticStaticDynamicCostLowModerateHighNo specific numerical values are presented here, since the types and characteristics of the strategies can vary depending on the context and conditions. This comparison aims to highlight the general trends rather than provide precise data. Comparison on different beam manipulation strategies No specific numerical values are presented here, since the types and characteristics of the strategies can vary depending on the context and conditions. This comparison aims to highlight the general trends rather than provide precise data. Lens designs generally offer a cost-effective solution for static beam manipulation, as they can be fabricated using commercially available 3-D printing technology or CNC machining. These printed lenses provide continuous phase profiles, ensuring accurate beam manipulation performance in most cases. Given that most materials can be considered as dispersionless within the communications bandwidths, the resulting phase profile is linearly proportional to the wavenumber and the lens designs are conceptually achromatic. A key limitation of lens designs is their large device size, which can lead to increased intrinsic material losses. Consequently, the aperture size of the lens design cannot be too large, and the overall compactness is compromised. Introducing phase wrapping enables a reduced device profile, but at the cost of a decreased bandwidth. To improve compactness, metasurfaces provide an alternative solution. They involve advanced fabrication techniques such as thin-film spin coating or electron beam deposition, making the fabrication process more complicated than lens devices. The unit cells of a metasurface are required to cover at least a 2π phase range, and a 3-bit configuration is typically sufficient. For example, eight unit cells are designed to cover the 2π phase range, with each unit cell contributing a phase shift of approximately 45∘. While more precise phase manipulation using higher-order bit configurations is possible, it introduces additional complexity. The required phase profile is discretized, and the unit cells are mapped to the phase profile accordingly. This results in relatively good beam manipulation performance, although discretization might slightly limit the precision. Due to the necessary phase wrapping and intrinsic dispersion where the phase difference between unit cells does not remain constant across a broad frequency band, the bandwidth performance is not as large as that of lens designs. However, it is typically sufficient for most communications applications. Metasurfaces are designed with tailored impedance matching, therefore minimizing radiation losses and ensuring good efficiency. A challenge with metasurfaces is that the amplitude response is also coupled to the phase response. For example, the amplitude responses of the eight unit cells covering the 2π phase range might be slightly different, which can lead to complexity in designing a particular wavefront. Nonetheless, there is no strict limitation on the aperture size, allowing for the design of very large metasurfaces. In summary, metasurfaces offer a more compact solution than lenses with acceptable performance, but they are still designed for one specific goals and do not offer reconfigurability. RISs are promising solutions for enabling reconfigurability, as the phase response of their unit cells can be dynamically adjusted, allowing for the realization of various beam manipulation goals. However, enabling such reconfigurability also introduces several constraints. Take semiconductor-based RIS as an example, the phase response is controlled by switching the active components on and off. As a result, the phase profile is typically discretized into a 1-bit stage, meaning the phase can only take values of 0∘ or 180∘. Consequently, the performance of RIS is generally not as good as lens or metasurface strategies. Furthermore, the phase difference between the on and off states of the active components can maintain at 180∘ for a narrow frequency range, limiting the flexibility of RIS to operate at broader frequencies. In the case of tunable material-based RIS, multi-bit phase quantization is also possible, with an improved bandwidth performance. However, these RISs face challenges such as slow switching speeds, often on the millisecond scale, which limits their practical applications. The efficiency of different RIS implementations is typically lower than that of lens or passive metasurface designs. Specifically, at the unit-cell level, active components are intrinsically less efficient at terahertz frequencies and introduce amplitude variations that can reduce the reflection efficiency. At the implementation level, biasing networks and interconnect structures introduce additional parasitic losses that contribute to efficiency reduction. At the system level, continuous power consumption by biasing and the control board is required to enable reconfigurability. As a consequence, scaling RIS to very large aperture sizes becomes challenging. Typically, for independently controlled unit cells, transistor-based RIS implementations are limited to apertures on the order of 10 × 10, while RISs based on tunable materials generally reach around 50 × 50 level. In contrast, CMOS-based RISs can support larger apertures through on-chip integration. For even larger arrays, practical implementations often adopt subarray-based architectures or shared-bias configurations, in which multiple unit cells (e.g., along a row or column) are driven by a common control line, trading off accuracy for scalability. In summary, RISs offer a compact solution for achieving reconfigurability in beam manipulation, but at the cost of reduced performance and higher complexity. Looking forward, the practical evolution of RIS is increasingly driven by three directions: reducing the reliance on densely distributed active components, lowering control and interconnect complexity, and emphasizing holistic system co-design rather than aggressive unit-cell-level reconfigurability. Despite these limitations, RISs play a crucial role in the future development of terahertz communication systems, especially as technologies for dynamic beam manipulation continue to evolve. Conclusion and outlook This article provides an overview of techniques for achieving highly efficient beam manipulation in the terahertz range. We laid out the fundamental principles of EM wave propagation, followed by detailed beam manipulation techniques that address various practical challenges in terahertz communications scenarios. The core principle of beam engineering, manipulating the complex electric field by adjusting the aperture’s amplitude and phase distribution to achieve desired beam behavior, is discussed in detail. Experimental validations using lens and metasurface solutions are presented. Alternative methods, such as RIS, are also briefly explored. The primary contribution of this article is to provide comprehensive guidance on beam manipulation for the terahertz communications community, detailing critical steps such as principles, design, fabrication, and measurement processes to realize cross far- and near-field coverage. Compared with existing review articles that often focus on specific beam types, device-level implementations, or metasurface perspectives, this work emphasizes a communications-driven view, in which beam manipulation is systematically linked to deployable terahertz communications scenarios. It is important to acknowledge that the beam manipulation techniques presented here remain constrained by the current state of fabrication and implementation technologies. While a variety of scenarios are explored, they are not exhaustive. Looking ahead, the advancement of solid-state terahertz sources with dynamic beam manipulation capabilities, potentially implemented through efficient phased-array antennas, will be critical for future terahertz systems. These systems will require increasingly complicated beam manipulation techniques, and the evolution of these techniques must align with the progress in fabrication technologies. Furthermore, future deployable terahertz communications systems are expected to operate in complex environments. In such scenarios, a key challenge lies not only in how to realize advanced beam manipulation, but also in how to determine which beam manipulation strategy is most suitable. This naturally calls for the assistance of sensing capabilities, enabling the system to perceive its surroundings and adaptively select or reconfigure beam patterns. From this perspective, beam manipulation and sensing are expected to evolve in a tightly coupled manner, jointly enabling robust and flexible terahertz communication systems. Supplementary information Transparent Peer Review file Transparent Peer Review file