Rotatable Antenna-Empowered Wireless Networks: A Tutorial

1 Rotatable An tenna-Emp o w ered Wireless Net w orks: A T utorial Beixiong Zheng, Senior Mem b er, IEEE, Qing jie W u, Xue Xiong, Y anhua T an, W eih ua Zhu, Tian tian Ma, Changsheng Y ou, Member, IEEE, Xiao dan Shao, Mem b er, IEEE, Lip eng Zh u, Senior Member, IEEE, Jie T ang, Senior Member, IEEE, Rob ert Sc hob er, F ellow, IEEE, Kai-Kit W ong, F ello w, IEEE, and R ui Zhang, F ellow, IEEE Abstract—Rotatable an tenna (RA) has emerged as a promising tec hnology for enhancing wireless communication and sensing p erformance through exible an tenna orienta- tion/b oresigh t rotation in the three-dimensional (3D) space. By enabling mec hanical or electronic b oresigh t adjustment without altering ph ysical an tenna p ositions, RA in tro duces additional spatial degrees of freedom (DoF s) b eyond con- v entional b eamforming based on xed antennas, oering a ligh tw eight and hardw are-ecient paradigm for an tenna ar- c hitecture design. In this pap er, we pro vide a comprehensive tutorial on the fundamentals, architectures, and applications of RA-emp o w ered wireless netw orks. Sp ecically , we begin by reviewing the historical ev olution of RA-related technologies and clarifying the distinctiv e role of RA among exible an tenna arc hitectures. Then, w e establish a unied mathematical frame- w ork for RA-enabled systems, including general antenna/arra y rotation mo dels that capture b oresight-dependent directional gain, as well as c hannel mo dels that co ver near- and far-eld propagation characteristics, wideband frequency selectivity , and p olarization eects. Building up on this foundation, we in- v estigate an tenna/array rotation optimization in represen tativ e comm unication and sensing scenarios for dierent p erformance ob jectiv es. F urthermore, we examine RA c hannel estima- tion/acquisition strategies encompassing orien tation sc heduling mec hanisms and signal pro cessing metho ds that exploit multi- view c hannel observ ations. Beyond theoretical mo deling and algorithmic design, we discuss practical RA congurations and deplo yment strategies, highlighting key design trade-os for hardw are implementation and system architectures. W e also presen t recent RA protot yp es and exp erimen tal results that v alidate the practical p erformance gains enabled by antenna Beixiong Zheng, Qing jie W u, Xue Xiong, Y anh ua T an, W eih ua Zhu, and Tiantian Ma are with the School of Micro electron- ics, South China Univ ersity of T ec hnology , Guangzhou 511442, China (e-mail: b xzheng@scut.edu.cn; miqjwu@mail.scut.edu.cn; ftx- uexiong@mail.scut.edu.cn; bctany anh ua06@mail.scut.edu.cn; miw ei- huazh u@mail.scut.edu.cn; mitiantianma@mail.scut.edu.cn). Changsheng Y ou is with the Department of Electronic and Elec- trical Engineering, Southern Universit y of Science and T ec hnology (SUST ech), Shenzhen 518055, China (e-mail: y oucs@sustec h.edu.cn). Xiaodan Shao is with the Department of Electrical and Com- puter Engineering, Univ ersity of W aterloo, W aterloo, ON N2L 3G1, Canada (e-mail: x6shao@uw aterloo.ca). Lipeng Zh u is with the State Key Laboratory of CNS/A TM and the School of Interdisciplinary Science, Beijing Institute of T echnology , Beijing 100081, China (e-mail: zh ulp@bit.edu.cn). Jie T ang is with the School of Electronic and Information Engi- neering, South China University of T ec hnology , Guangzhou 510640, China (e-mail: eejtang@scut.edu.cn). Robert Schober is with the Institute for Digital Comm uni- cations, F riedric h-Alexander-Universit y Erlangen-N ¨ u rnberg (F AU), 91054 Erlangen, Germany (e-mail: rob ert.sc hob er@fau.de). Kai-Kit W ong is with the Department of Electronic and Electrical Engineering, Universit y College London, WC1E 7JE London, U.K. He is also with Y onsei F ron tier Lab, Y onsei Universit y , Seoul 03722, South Korea (e-mail: kai-kit.wong@ucl.ac.uk). Rui Zhang is with the Department of Electrical and Computer Engineering, National Universit y of Singapore, Singap ore 117583 (e- mail: elezhang@nus.edu.sg). rotation. Finally , w e highlight promising extensions of RA to emerging wireless paradigms and outline op en challenges to inspire future research. Index T erms—Rotatable an tenna (RA), antenna orienta- tion/b oresigh t, wireless communication/sensing, rotation op- timization, channel estimation, RA architecture, 6G. I. In tro duction A. Background and Motiv ation Ov er the past few decades, wireless communication tec hnologies hav e adv anced at an unprecedented pace, profoundly reshaping human life and mo dern so ciet y [ 1 ]. Bey ond merely facilitating interpersonal connectivity , wireless netw orks ha ve evolv ed into critical infrastructure that underpins economic developmen t, supp orts public services, and safeguards national security . As the fth- generation (5G) netw ork reaches global maturit y , the re- searc h communit y is activ ely shaping the sixth-generation (6G) era [ 2 ]. F uture 6G netw orks are en visioned to supp ort a set of representativ e usage scenarios identied in the International Mobile T elecommunications (IMT)- 2030 framework, including immersiv e communication, h yp er-reliable and low-latency comm unication (HRLLC), massiv e communication, integrated articial intelligence (AI) and communication, ubiquitous connectivit y , and in tegrated sensing and communication (ISA C) [ 3 ]. These emerging scenarios imp ose stringent p erformance require- men ts, including ultra-high data rates, robust reliability , ubiquitous cov erage, ultra-low latency , as well as inte- grated in telligence and sensing capabilities, which p ose signican t challenges to curren t wireless systems [ 4 ]–[ 8 ]. T o accommo date these increasingly am bitious demands, the evolution of wireless net works from generation to generation has primarily follow ed tw o technical pathw ays: expanding system bandwidth in the frequency domain and increasing the num b er of antennas to exploit additional degrees of freedom (DoF s) in the spatial domain. How- ev er, sp ectrum resources are becoming increasingly scarce, particularly in the commercially lo w-frequency bands. Due to the limited av ailability of fragmented frequency sp ectrum, further expanding bandwidth has b ecome chal- lenging. Although shifting to higher-frequency bands (e.g., millimeter-w av e (mm W av e) and terahertz (THz)) oers wider bandwidths, this is accompanied by m uch higher path loss, increased susceptibility to blo ckage, and more stringen t hardw are design requirements. Consequen tly , deplo ying a larger n umber of antennas at the base station 2 (BS) is b eing pursued to enhance sp ectral eciency in cellular netw orks [9]. This evolution has driven m ultiple- input multiple-output (MIMO) technology tow ard more adv anced architectures, such as massive MIMO [10], [11] and extremely large-scale MIMO (XL-MIMO) [12], [13]. By deplo ying h undreds or ev en thousands of antennas at the BS, these arc hitectures can ac hiev e signican t spatial m ultiplexing and b eamforming gains, thereb y comp ensat- ing for the limitations in spectral resources. Despite their p oten tial, these an tenna-scaling ap- proac hes are increasingly limited by hardw are complex- it y and energy consumption. The realization of massive MIMO and XL-MIMO systems entails substantial c hal- lenges, including high radio-frequency (RF) c hain cost, increased circuit pow er consumption, and considerable signal pro cessing complexity . F urthermore, the b enets of simply increasing the n um b er of antennas diminish rapidly due to the la w of diminishing returns, where the marginal p erformance gain fails to oset the substantial increase in hardware complexity and cost [ 14]. Although v arious more cost-eective alternatives such as sparse arra ys [15], lens antenna arrays [16 ], and intelligen t reecting surfaces (IRSs) [17]–[22] hav e b een in vestigated to mitigate these issues, most of them still rely on the con ven tional xed- an tenna architecture. In xed-antenna systems, b oth the p osition and orien tation of the antennas remain static p ost-deplo ymen t, resulting in an inherent mismatch b e- t ween the ph ysical transceiv er and the dynamic wireless propagation en vironment. This lack of antenna adaptabil- it y preven ts existing systems from fully exploiting the rich spatial DoF s av ailable in dynamic wireless environmen ts, often leading to degraded cov erage and limited capacity in complex scenarios. T o o vercome these in trinsic limitations, there is a growing need to mov e b ey ond xed antenna designs tow ard exible architectures that can actively adapt to the propagation environmen t and make better use of antenna resources. B. What Is RA and How It W orks Motiv ated by the ab ov e, rotatable antenna (RA) has recen tly emerged as an ecient tec hnology to enhance wireless communication and sensing p erformance through adaptiv e control of antenna orientation/boresight [23]– [26]. In particular, by allowing eac h antenna to indep en- den tly rotate its orientation/boresight tow ard a desired direction in three-dimensional (3D) space, RA in tro duces additional spatial DoF s b ey ond conv entional b eamform- ing based on xed antennas. This capabilit y allows the radiated energy to b e not only conned within a narrow angular region but also physically steered to ward a specic spatial p oint, forming a directional “sp otligh t” eect that enhances the eective array gain tow ard intended users or targets, as illustrated in Fig. 1 . Additionally , the rotational capabilit y also enables RA to scan the 3D space in an eye- lik e manner, thereby ac hieving broader spatial co verage for b oth communication and sensing. Moreov er, through ne-grained orien tation/b oresight control, RA enhances the eective directional gain tow ard desired users or targets while suppressing radiation in undesired directions, thereb y improving p o w er eciency and system p erfor- mance with signicantly few er RF c hains than traditional xed-an tenna systems. In practical implemen tations, RA designs can b e ac hieved through mec hanical control or electronic con- trol, as illustrated in Fig. 1. Specically , mec hanical con trol approaches typically utilize serv o motors or micro- electromec hanical systems (MEMS) to physically rotate the orien tation of directional antennas, oering a wide angular adjustmen t range with milliwatt-lev el p o w er con- sumption and millisecond (ms)-scale resp onse times [27], [28]. In contrast, electronic control metho ds retain a xed an tenna orientation while enabling rapid b oresigh t adjustmen t through electronic tec hniques suc h as Positiv e- In trinsic-Negative (PIN) diode switching or recongurable parasitic radiator element loading, thus achieving m uch faster resp onse times at the microsecond (µs)/nanosecond (ns) scale and b etter compatibility with existing wire- less systems [29]–[31]. T o harness the b enets of b oth approac hes, co-designed RA architectures that in tegrate b oth driving mechanisms can also b e adopted to en- able wide-angle and rapid rotation of an tenna orienta- tion/b oresigh t. Through these capabilities, RA achiev es notable p er- formance gains in wireless communication systems, par- ticularly in terms of adaptive in terference circumv ention, b eam fo cusing, and cov erage extension, as illustrated in Fig. 2 . F or instance, as the in terference typically comes from sp ecic directions, RA can physically rotate its an tenna b oresigh t aw ay from those directions to av oid in terference, thereby reducing reliance on sophisticated signal pro cessing techniques. In conv entional systems, b eamforming is typically realized at the array lev el b y applying complex-v alued weigh ts to iden tical transmitted sym b ols across multiple antennas, enabling directional transmission through coheren t sup erp osition and cancel- lation of radiated signals (see Fig. 1 ). Building up on con ven tional b eamforming, RA arra ys can further reorient eac h individual an tenna’s radiation p o wer distribution to ac hieve sharp er b eam fo cusing. F urthermore, RA arrays can dynamically adjust antenna orientations to exibly extend the eectiv e comm unication co verage across a m uch wider 3D space without increasing the transmit p o w er or requiring denser infrastructure deploymen t. This capabilit y allows cov erage of users and targets distributed across dierent altitudes and spatial regions, which is particularly desirable in low-altitude ISAC and space-air- ground in tegrated net works (SAGIN). Moreo ver, in hybrid congurations where RAs are co-deploy ed with xed- sector antenna arrays, the exible reorientation capability of RAs can complement the static co verage of xed- sector antennas b y eectively mitigating blind sp ots and enhancing link reliabilit y , esp ecially in distributed or irregular deplo yment scenarios. In parallel, RA-enabled sensing systems also demon- strate signicant p erformance adv an tages compared to 3 RA controller RA array RA elem e n t … … … … RA Arra y System θ Si g n a l Beamforming gain Boresight alignment gain Fewer RF chains Beamforming gain Si g n a l AI R F co nt rol li nk R otat io n con tr ol l in k Radio Sensing Vision Traditional Array RA Array BCB hinge Subs trat e Ground Microstrip lin e Rotational axis Patch Biasing layer S elect d iff erent ant enn a feed combinations Antenna feed Electrically - driven RA Yaw axis Pit ch axis Mechanically - driven RA MEMS - based RA Pan - tilt head - based RA Fig. 1. F rom conv entional xed-antenna array to RA array: Arc hitecture and implementations. In t e rfere n ce … … … … A i r s p ace G ro u n d Mu l t i p l e t arg et s … … … Spee d Dire c ti on S ha pe … … … S pe c ific t a rge t Interference c irc um vent i on Beam foc using Coverage ext ension High - resolution s e ns i ng Multi - dimensional sensing Multi - target sensing (a) Communication (b) Sensing Fig. 2. F unctional adv antages of RA for comm unication and sensing. con ven tional xed-antenna systems, owing to their unique b oresigh t trac king/scanning capabilit y , as illustrated in Fig. 2 . Ev en with a v ery limited n umber of an tennas, RA arrays can supp ort ecient multi-target sensing by exploiting the an tenna rotation and spatial scanning capabilities to sense and detect targets from div erse spatial directions, allowing the num b er of detectable targets to signicantly exceed the num b er of an tennas. Bey ond wide-area scanning for detecting multiple targets, RA arrays can further enable high-resolution sensing by coheren tly rotating the b oresigh t directions of multiple RAs to ward a common spatial lo cation. Building on high- resolution sensing, exible boresight con trol of RAs further enables multi-dimensional sensing by facilitating more eectiv e extraction of ric h target features. By adaptively trac king targets from dierent orientations, RA-enabled sensing systems can capture additional information such as target velocity , mov ement direction, shape, and other ne-grained c haracteristics with higher reliability . This capabilit y is particularly b enecial for dynamic sensing scenarios inv olving mobile or clustered targets, where con- v entional xed-antenna systems often suer from limited angular div ersity and incomplete feature observ ation. C. Historical Dev elopment and F uture T rend of RA- Related T ec hnologies F rom early concepts of mechanically rotating an tennas used for direction nding to the modern realization of exible an tenna architectures, impro ving the eciency and controllabilit y of wireless transmission has remained a central theme [32]. A foundational milestone in this ev olution o ccurred in 1901, when Guglielmo Marconi’s transatlan tic radio exp erimen t marked the b eginning of long-distance wireless communication and implicitly high- ligh ted the critical role of spatial radiation control [33]. Subsequen tly , v arious directional antennas, including lo op an tennas, Y agi-Uda antennas, and parab olic reectors, w ere dev elop ed and subsequen tly deplo yed across com- m unication, radar, and radio astronomy applications [34]. These antennas were commonly installed on mechanically 4 rotatable platforms that allow ed dynamic adjustment of their radiation directions, representing an early precursor to the mo dern RA arc hitectures. Building on this de- sign paradigm, practical implementations of mechanical an tenna rotation were widely utilized during W orld W ar I I [35], [36]. Particularly , by using a carrier-based rotating an tenna to transmit unique Morse co de letters across t welv e 30-degree sectors, the U.S. Na vy successfully ad- dressed the critical challenge of precision navigation ov er the featureless op en o cean [35]. F urther miniaturization of this technology was achiev ed through adv ances in MEMS at the end of the tw entieth century , yielding more compact and ecient antenna rotation mechanisms with fast, low-pow er, and more exible mec hanical rotation capabilit y [27], [37], [38]. Despite its eectiveness, mechanical antenna rotation b ecame increasingly inadequate as target (e.g., aircraft) sp eeds increased, thereby motiv ating the developmen t of electronic radiation b eam con trol. Initially , phased- arra y an tennas w ere used to direct beams to ward desired directions b y imp osing con trolled phase dierences across an tenna elements [34]. How ever, such systems require a large num b er of phase shifters, which incur substantial cost particularly at higher frequency bands. In addition, the radiation eciency of phased-array antennas degrades at large steering angles. In this context, an alternativ e approac h that has gained prominence is the reactively con trolled directiv e arra y , in which the main b eam is electronically steered by appropriately selecting reactiv e loads, thereb y providing a ligh tw eight and structurally simplied alternative to conv entional phased arrays [39], [40]. Around the early 2000s, a ma jor step tow ard practical electronic b eam con trol was enabled by the adoption of semiconductor devices such as PIN dio des and v aractor dio des. By manipulating the current ows on the an tenna through these RF switc hing dio des, the an tenna radiation pattern can b e accordingly mo died [41]. Moreov er, adv ances in pixel antenna tec hnology furt her accelerated the developmen t of electronic b eam control. By con- trolling the PIN dio des connecting the pixels, pixel an- tenna systems oer larger b eam rotation cov erage, higher scanning resolution, and more recongurable radiation patterns [29]–[31]. F or example, a recen tly proposed planar pixel antenna enables full 360-degree 3D b eam scanning capabilit y with v ery low gain uctuation, closely aligning with the electronically controlled RA concept [29]. While exible antenna architectures remained largely application-sp ecic in the past, the increasing demand for higher spatial resolution and more adaptive cov erage in adv anced wireless systems has renewed attention on exible antenna technologies recen tly [42]. In this con text, the notion of “mo v able an tenna (MA)” w as formally in tro- duced in 2007 [43] and applied to wireless comm unications in 2009 [44]. In parallel, while the term “uid antenna” w as rst in tro duced in [45] and traditionally corresponds to liquid-based radiators [46], [47], W ong et al. introduced the concept of Fluid An tenna System (F AS) in 2020 [48]– [50] to advocate softw are-controlled p osition- and shap e- exible antennas capable of exploiting spatial diversit y . Precisely , dieren t from b eing an antenna terminology , F AS is not an an tenna technology p er se but a hardware- agnostic system concept that treats the antenna as a recongurable physical lay er resource to broaden system design, encompassing softw are-controllable uidic, con- ductiv e, and dielectric structures capable of reconguring fundamen tal an tenna c haracteristics suc h as gain, radia- tion pattern, and op erating frequency [51]–[55]. Both MA and F AS share the same concept of antenna p ositional recongurabilit y to enrich the eective spatial diversit y of wireless c hannels, while diering in their underlying hardw are. Building up on these adv ances, a joint transmit- receiv e MA-aided wireless communication system with t wo-dimensional (2D) an tenna mov ement was prop osed in 2022 [56], [57], follo wed by its extension to 3D mo vemen t in [58]. T o extend spatial recongurability b ey ond exist- ing MA and F AS designs, the six-dimensional mov able an tenna (6DMA) framework was prop osed in 2024 [59]– [61], incorp orating b oth 3D p ositional and rotational con trol, thereby achieving enhanced exibility without increasing the num b er of antennas. Moreov er, pinc hing an tenna, rst introduced in 2022 [62], can b e conceptually traced back to earlier visions of programmable surface- w av e path wa ys with con trollable leaky radiation [63], an application of p osition-recongurable F AS in the large scale. This tec hnology provides another path to large-scale spatial adaptation by radiating from arbitrary p oin ts along dielectric w av eguides [64], [65]. Driv en by the gro wing interest in spatially adaptive systems, RA technology emerged in late 2024-2025 as a rotation-cen tric exible antenna architecture that en- ables spatial adaptability through controllable 3D antenna orien tation/b oresigh t rotation [23]–[25]. In particular, in- spired by the ph ysical antenna motion in MA/6DMA, mec hanically-driven RA w as prop osed to ac hieve spatial adaptabilit y by physically rotating the antenna orienta- tion. Mean while, motiv ated by the in ternal reconguration mec hanism of F AS based on electronic switc hing elements, electronically-driv en RA w as developed to reshap e the radiation pattern b y switching the radiation main-lobe direction (without ph ysically c hanging an tenna orien- tation). T o clearly highlight the key characteristics of dieren t exible antenna arc hitectures, T able I provides a comparativ e summary of typical designs, including F AS, MA, 6DMA, pinching-an tenna systems (P ASS), and RA. In some sense, RA can also b e viewed as a simplied y et promising mem b er of the F AS/MA/6DMA family that retains only the antenna rotation exibilit y (without changing antenna p osition), thereby oering a cost-eectiv e and compact solution. Compared with translational motion in MA and 6DMA, which requires additional ph ysical space (e.g., sliding tracks or mov able regions), RA ac hieves recongurability through localized rotation, allowing more eciently integrated mec hanical or electronic implementations. By conning the recong- uration to the orien tation/b oresigh t control, RA strikes a fa vorable balance betw een exibility and complexit y , 5 T ABLE I Comparison of Dierent Flexible Antenna Arc hitectures An tenna Arc hitecture Hardw are Implementation Recongurable Parameter Deplo yment Complexit y System Ov erhead RA [23]–[25] Mec hanical or electronic means An tenna 3D b oresigh t direction Lo w Lo w F AS [48]–[50] Pixel antennas or liquid materials Antenna shap e and 3D p osition Mo derate Mo derate MA [56]–[58] Motors An tenna 3D p osition Mo derate Mo derate 6DMA [59]–[61] Motors and exible cables An tenna 3D p osition and 3D rotation Mo derate to high Moderate to high P ASS [62], [64] W av eguides and distributed pinc hing antennas Large-scale antenna p osition Mo derate to high Moderate to high 1901 Marconi ’ s transatlantic radio experiment with a kite-trailed antenna [33] Emergence of various directional antennas with rotatable platforms [34] RA-Related Technologies World War II Widespread use of rotating antennas in mechanically scanned radars and ocean navigation [35], [36] 2000 1945 Emergence of semiconductor devices enabled electronically controlled antennas [41] 1939 Emergence of MEMS-enabled antennas with fast mechanical rotation control capability [27], [37], [38] Emergence of reactively controlled directive array [39] 1978 Introduction of pixel antenna technologies to enhance the beam rotation capability [29]-[31] First mentioning of “movable antenna” [43] First mentioning of “fluid antenna” [45] 2020 Redefinition of “fluid antenna system” (FAS) and analysis of wireless communication performance [48]-[50] 2022 First mentioning of “pinching antenna” [62] 2024 Introduction of 6DMA with flexible antenna position and rotation and application to wireless communications [59]-[61] 2025 Introduction of rotatable antenna (RA) and derivation of channel model and optimization framework for RA systems [23], [24] 2007 2008 Year 2004 First appearance of “liquid antenna” [47] Channel model and optimization framework for movable antenna (MA) systems [56], [57] Introduction of pinching antenna to wireless communications [64], [65] 2009 Introduction of movable antenna in wireless systems [44] Fig. 3. Illustration of the historical developmen t of RA and other exible antenna architectures. oering a ligh tw eight yet eective solution for directional con trol, p olarization alignment, and spatial adaptability in next-generation wireless systems. T o provide a clearer o verview, the key milestones in the historical developmen t of RA-related technologies are summarized in Fig. 3 . Lo oking ahead, RA technology is exp ected to play an increasingly imp ortan t role in compact and energy- ecien t wireless comm unication and sensing systems. In particular, the conv entional 2D ground co verage en- abled by xed-sector antennas is no longer sucien t to accommo date the growing demands of emerging 3D connectivit y and p erception scenarios inv olving aerial and non-terrestrial no des, such as in the low-altitude economy (LAE) and SA GIN. With lo w-complexity b oresigh t control and ne-grained directional adaptability , RA enables a paradigm shift from static sector-based 2D cov erage to exible 3D spatial transmission across diverse deploymen t en vironments. As illustrated in Fig. 4 , RA arrays can b e deploy ed across space, air, and ground segments, allo wing wireless infrastructures to dynamically adapt their radiation directions in resp onse to spatially dis- tributed and dynamic service demands. Bey ond enhancing con ven tional urban and suburban outdo or cov erage, RA is particularly attractiv e in scenarios where dense BS deplo yments are infeasible or inecient. F or example, in deserts, mountainous regions, and o cean environmen ts, RA-equipp ed aerial platforms or satellites can dynamically rotate antenna boresight directions to pro vide scalable co verage and reliable backhaul links o ver sparse, mobile, or infrastructure-limited netw orks. In more challenging envi- ronmen ts suc h as underwater and deep-o cean operations, where terrestrial connectivit y cannot b e established, RA- enabled maritime or submarine platforms oer a promising means to maintain directional, robust, and task-oriented comm unication and sensing links. In addition to large-scale outdo or deplo yments, RA also exhibits signicant p oten tial in indo or and short- range scenarios that demand high spatial selectivity and exible cov erage adaptation. In indoor environmen ts suc h as smart factories, smart oces, and residential spaces, RAs can dynamically adjust their b oresigh t directions to align with the locations of active users, devices, or mac hines, thereby impro ving link reliability , mitigating in terference, and enhancing energy eciency . Moreov er, in emerging immersive applications such as virtual reality (VR), augmented reality (AR), and indo or digital twin systems, RA enables more precise spatial focusing of signal energy , supp orting high-data-rate, lo w-latency commu- nications and high-accuracy sensing within conned 3D spaces. Bey ond traditional static deplo yment scenarios, RA can further benet emerging application scenarios in volving mobile platforms and dynamic en vironments, suc h as intelligen t transp ortation systems, high-sp eed v ehicles, and industrial rob otics, where rapid changes in relativ e geometry p ose signicant challenges to xed- an tenna solutions. By con tinuously adapting an tenna orien tation/b oresigh t in response to en vironmental and mobilit y v ariations, RA oers a practical means to sustain 6 atellite network RA AR/VR Smart Factory Smart Office ISAC Network Smart Transportation Smart Home/Factory LEO Satellite Com m unicati on Satellite Network SAGIN Env ir onme nt al mo nitor ing Emergency Communication Ultr a - high data r ate supp or t M as s i ve co nnect ion Indoor localization Shore - based Communication RA - aided navigation Rotatable tra ns d uce r Undersea Network C omm unica tion Sig nal Sen si n g Si gnal Be am Focusi ng/ Tr ackin g R otata ble A nt en na ( RA) Fig. 4. Application scenarios of RA-enabled wireless netw orks. directional links and stable p erformance under highly dy- namic conditions. Mean while, suc h directional agility also supp orts accurate sensing, lo calization, and na vigation b y adaptively aligning the radiation main-lob e direction to ward mobile targets or dynamic areas of interest. These adv antages highligh t the practical scalability of RA across heterogeneous platforms and ev olving netw ork top ologies, la ying a foundation for its broader in tegration into future wireless systems. D. Ob jective, Contribution, and Organization Giv en the signicant p oten tial of RA for unlo c king new DoF s with lo w hardware complexity , this pap er pro vides a comprehensive tutorial on the fundamentals, implemen tation, and recent adv ances of RA-emp o w ered wireless netw orks. Specically , it cov ers the mathemat- ical framew ork for an tenna/array rotation and channel mo deling, the main design challenges for RA-enabled systems, hardw are implemen tation arc hitectures, pro of-of- concept protot yp es, and emerging application scenarios. Our ob jective is to establish a solid theoretical and practical foundation for researchers in this eld and to inspire future inv estigations. Although exible antenna architectures ha ve recently attracted extensive attention, existing works are typically dev elop ed based on sp ecic reconguration dimensions and their associated mo deling assumptions, whic h are not directly transferable to RA systems. In particular, researc h on MA/F AS has mainly inv estigated p osition reconguration within a prescrib ed region, where p erfor- mance impro vemen ts are realized through translational mo vemen t and the resulting p osition-dep enden t channel v ariation [51], [66]–[69]. Mean while, most existing 6DMA- related w orks assume simplied signal models, whic h ov er- lo ok practical asp ects such as wideband frequency selec- tivit y and p olarization eects [61], [ 70]. More imp ortan tly , existing RA-related works are largely limited to conceptual discussions or sp ecic optimization problems, and th us do not yet provide a comprehensive introduction that bridges the gap b etw een theoretical modeling and practical implemen tation. In addition, although several magazine pap ers [ 25], [26], [71] hav e discussed the basic princi- ples, k ey c hallenges, and opp ortunities of RA systems, they generally lack comprehensiveness and comparisons of dieren t channel mo deling methods, channel estimation strategies, detailed optimization approaches, and state-of- the-art prototypes. Compared with these existing w orks, the main contributions of this tutorial can b e summarized as follo ws: • W e pro vide the motiv ation and historical dev elop- men t for RA-emp o w ered wireless netw orks and es- tablish a unied fundamental framew ork for RA- enabled systems, including a general antenna/arra y rotation mo del and representativ e c hannel mo dels that capture b oresight-dependent directional gain as w ell as polarization eects. • W e inv estigate antenna/arra y rotation optimization in represen tative RA-aided systems, co vering RA for comm unication scenarios (e.g., single- input single-output (SISO)/multiple-input single- output (MISO)/single-input m ultiple-output (SIMO)/MIMO, single-user/m ulti-user, and 7 I. Introduc tio n A. B a c kgr ound a nd M oti va ti on B . W ha t I s R A a nd How It W orks C . Hist or i c a l De ve lopm e nt and Fut ure Tr e nd of R A - R e late d T e c hnologi e s D. O bje c ti ve , C ont r ibut ion, and O rga niz a ti on I I . R A F unda m enta l s A. A nte nna /Arr a y R ot a ti on Mode l B . C ha nne l Model C . Optim i z a ti on F r a mew or k a nd De si gn I ssues I I I . A nten na/A rra y R o ta ti o n Opt i m i z a ti o n A. RA - Ena ble d MIS O / S IMO S ystem B . R A - Ena ble d MIMO Syst e m C . RA - Ena ble d Mult i - Use r S yst e m D. RA - Ena ble d W ideba nd S ystem E. RA - Ena ble d ISAC S ystem I V . R A C ha nnel Es ti m a ti o n/ A cqui s i ti o n A. R A Orie nt a ti on S c he duli ng f or Channe l Est im a ti on B . C ha nne l Est im a ti on fo r D iff e re nt R A S ystem S e tups C . S i gna l P roc e ssi ng Me thods for R A Cha nne l Est im a ti on VI . R A P ro to ty pes a nd R el a te d Pr o duc ts V I I . Ex tens i o ns a nd F uture Direc ti o ns A. S i ngle RA P rotot ype s B . R A A rr a y P rototype C. Re la te d Comme r c ia l P r oduc ts A. Low - Alti t ude ISAC B . C og ni t i ve R a di o ( C R ) S ys t ems C . P hysi c a l La ye r Sec ur it y D. C e ll - F re e MIMO N e t w orks E. S i mul tane ous W ire less I nfor mation and Powe r T ra nsfe r (SW I P T) F . Othe r Mi sc e ll a ne ous Topic s V . R A C o nfi g ur a ti o ns a nd D epl o y m ent V I I I . C o n cl us i o ns A. Me c ha nica l vs. Ele c tronic R otation B . C onti nuous vs. Discr e te Rot a ti on C . S pa rse vs. Non - S pa rse Ar ra y D. Dist ribute d vs. C e ntra li z e d De ploym e nt Fig. 5. Organization of this pap er. wideband systems) and RA for ISAC scenarios for dieren t performance metrics. • W e presen t RA c hannel estimation/acquisition strategies including xed-orientation estimation and dynamic-orien tation estimation, and inv estigate rep- resen tative signal processing metho ds that exploit m ulti-view observ ations enabled by RA orientation con trol. • W e elab orate practical RA congurations spanning from concept to implemen tation, compare representa- tiv e conguration options (e.g., sparse/non-sparse ar- ra y structures, contin uous/discrete rotation, and dis- tributed/cen tralized deploymen t), and provide guide- lines for selecting suitable RA congurations under hardw are and con trol constrain ts. • W e provide an ov erview of RA prototypes and re- lated pro ducts, summarizing representativ e imple- men tation strategies and experimental results that v alidate the p erformance gains achiev able in practice through antenna rotation. Moreov er, we discuss op en c hallenges and future research directions to broaden the application scope of RA-enabled wireless systems. The organization of this pap er is illustrated in Fig. 5 . In particular, Section I in tro duces the motiv ation and back- ground of RA-emp ow ered wireless net works and reviews the historical dev elopmen t and future trends of RA-related tec hnologies. Section II presents the fundamen tals of RA systems, including the antenna/arra y rotation mo del and representativ e c hannel mo deling metho ds. Section I II in vestigates antenna/arra y rotation optimization in rep- resen tative RA-aided communication and ISA C systems. Section IV reviews RA channel estimation/acquisition strategies. Section V discusses RA congurations and deplo yments. Section VI presents pro of-of-concept pro- tot yp es and related commercial pro ducts. Section VI I pro vides extensions and future directions. Finally , this pap er is concluded in Section VI I I. Notation: Upp er-case and low er-case b oldface letters denote matrices and column vectors, resp ectively . ( · ) T , ( · ) ∗ , ( · ) H , and ( · ) − 1 stand for the transp ose, conjugate, Hermitian transp ose, and matrix in version op erations, resp ectiv ely . The sets of a × b dimensional complex and real matrices are denoted by C a × b and R a × b , resp ectiv ely . b·c is the o or function, ⊗ denotes the Kroneck er pro duct, and E {·} denotes the exp ectation of a random v ariable. F or a vector x , k x k denotes its ℓ 2 -norm, diag( x ) returns a diagonal matrix with the elements in x on its main diagonal, <{ x } and ={ x } denote its real and imaginary parts, respectively , and [ x ] i denotes its i -th entry . F or a matrix X , T r( X ) , and det( X ) denote its trace and determinan t, resp ectively , v ec( X ) is the vectorization op erator applied to X , and X  0 implies that X is p ositiv e semi-denite. I and 0 denote an identit y matrix and an all-zero matrix, resp ectiv ely , with appropriate dimensions. The distribution of a circularly symmetric complex Gaussian (CSCG) random v ector with zero mean and cov ariance matrix Σ is denoted by N c ( 0 , Σ ) ; and ∼ stands for “distributed as” . I I. RA F undamentals This section introduces the fundamentals of RA-enabled wireless systems. W e rst present the antenna and ar- ra y rotation mo dels, which capture the exible orien ta- tion/b oresigh t control of individual an tennas [23], [24] as well as the coordinated rotation of antenna arra ys. Next, we establish a near-eld channel model tailored for RA systems, which characterizes ho w antenna rotation mo dies the directional gain pattern observed b y wireless c hannels. Building on this foundation, the RA framew ork is further extended to co ver far-eld propagation, multi- path environmen ts, wideband scenarios, and p olarization eects. Finally , we dev elop a unied optimization frame- w ork for designing antenna orientations/boresights to enhance system performance. Unless otherwise sp ecied, w e consider a general system mo del comprising K single- an tenna users and a BS equipped with an RA array consisting of N directional an tennas for ease of exp osition. F or notational conv enience, we use subscripts “B”, “U”, and “C” to indicate the BS, user, and scatterer cluster, resp ectiv ely . A. Antenna/Arra y Rotation Model F or directional antennas/arra ys, the c hannel conditions dep end strongly on the relative geometric relationships 8 2D rotation Initial p osition 3D rotation z y RA n , n   f x Yaw y  () z z  y x x   Roll z z  () x x  y y   , zn  , an    f ⊥  f ⊥  f   f , n ⊥  f (b) (c) (d) (e) Pitch () y y  x z z  x   ⊥  f   f (a) z y x Fig. 6. Illustration of 3D and 2D rotations of an individual antenna. b et w een transceiv ers. In this regard, an tenna/arra y ro- tation alters b oth the directional gain to ward a spatial p oin t and the p olarization characteristics of the transmit- ted/receiv ed signal. According to Euler’s rotation theorem, an y 3D rotation around a xed p oin t can b e represen ted as a comp osition of three elemen tary rotations [72]. As illustrated in Fig. 6, the orientation of an an tenna/array can b e describ ed by three rotation angles: a roll angle ϕ ∈ [0 , 2 π ) around x -axis, a pitch angle θ ∈ [0 , 2 π ) around y -axis, and a ya w angle ψ ∈ [0 , 2 π ) around z -axis. W e adopt the x - y - z rotation sequence, with eac h rotation follo wing the righ t-hand rule. T o describ e the 3D orientation change of the an tenna in the global coordinate system, we adopt an extrinsic rotation description in which the an tenna rotates around the xed global co ordinate axes throughout the rotation pro cess. Let θ ≜ [ ϕ, θ , ψ ] T denote the rotation angle v ector. The corresp onding rotation matrix that transforms the an tenna’s lo cal co ordinate system (i.e., ˜ o - ˜ x ˜ y ˜ z ) to the global co ordinate system (i.e., o - xy z ) can b e expressed as [73]–[75] R ( θ ) = R z ( ψ ) R y ( θ ) R x ( ϕ ) =   c ψ − s ψ 0 s ψ c ψ 0 0 0 1     c θ 0 s θ 0 1 0 − s θ 0 c θ     1 0 0 0 c ϕ − s ϕ 0 s ϕ c ϕ   =   c θ c ψ s ϕ s θ c ψ − c ϕ s ψ c ϕ s θ c ψ + s ϕ s ψ c θ s ψ s ϕ s θ s ψ + c ϕ c ψ c ϕ s θ s ψ − s ϕ c ψ − s θ s ϕ c θ c ϕ c θ   , (1) where we use c x ≜ cos( x ) and s x ≜ sin( x ) for notational simplicit y . 1) Antenna Rotation: F or an RA, the antenna can indep enden tly rotate in 3D space while k eeping its position xed. The 3D spatial orien tation of RA n , n ∈ N ≜ { 1 , 2 , . . . , N } , can be uniquely determined by tw o non- parallel bo dy-xed v ectors [76]: • Poin ting vector  f ⊥ ,n ∈ R 3 × 1 , denoting the antenna b oresigh t direction (t ypically normal to the antenna line/plane); • Reference vector  f ∥ ,n ∈ R 3 × 1 , lying in the antenna line/plane and p oin ting to a xed azimuth reference. 3D Antenna Rotation: In the 3D Cartesian co ordinate system (CCS) shown in Fig. 6 (a), w e initialize the pointing and reference vectors of eac h antenna to b e parallel to the p ositiv e x - and z -axes, resp ectiv ely , i.e.,  f (0) ⊥ ,n = e 1 and  f (0) ∥ ,n = e 3 , where e 1 ≜ [1 , 0 , 0] T and e 3 ≜ [0 , 0 , 1] T . Given the rotation angle v ector θ n ≜ [ ϕ n , θ n , ψ n ] T , the p oin ting and reference vectors b ecome  f ⊥ ,n = R ( θ n )  f (0) ⊥ ,n =   c θ n c ψ n c θ n s ψ n − s θ n   , (2)  f ∥ ,n = R ( θ n )  f (0) ∥ ,n =   c ϕ n s θ n c ψ n + s ϕ n s ψ n c ϕ n s θ n s ψ n − s ϕ n c ψ n c ϕ n c θ n   , (3) where w e hav e k  f ⊥ ,n k = 1 and k  f ∥ ,n k = 1 due to normalization, and  f ⊥ ,n is p erp endicular to  f ∥ ,n , i.e.,  f T ⊥ ,n  f ∥ ,n = 0 . 2D An tenna Rotation: When the antenna directional gain pattern is rotationally symmetric, rotation around the b oresigh t axis do es not change the pattern. In this case, the directional gain depends primarily on the angular separation b etw een the signal propagation direction and the an tenna b oresigh t (i.e., the maxim um-gain direction of the mainlob e). The ab ov e 3D antenna rotation mo del can b e simplied b y omitting self-rotation around the b oresigh t axis, i.e., ϕ n = 0 , ∀ n ∈ N . As a result, the p oin ting and reference v ectors for the 2D an tenna rotation are giv en b y  f ⊥ ,n =   c θ n c ψ n c θ n s ψ n − s θ n   ,  f ∥ ,n =   s θ n c ψ n s θ n s ψ n c θ n   . (4) Alternativ ely , for ease of represen ting the pointing and reference vectors in the global coordinate system, they can also b e parameterized by a zenith angle θ z ,n and an azim uth angle θ a ,n , as in [23], [24]:  f ⊥ ,n =   c θ z ,n s θ z ,n c θ a ,n s θ z ,n s θ a ,n   ,  f ∥ ,n =   − s θ z ,n c θ z ,n c θ a ,n c θ z ,n s θ a ,n   . (5) As shown in Fig. 6 (e), the zenith angle θ z ,n denotes the angle betw een the p ointing vector and the x -axis, and the azimuth angle θ a ,n denotes the angle b et ween the pro jection of the p oin ting vector onto the y - z plane and the y -axis. 9 (a) Rotatable array (a rr a y - wise rotation) (b) Rotatable a ntenna and array … … Array - wise rotati on Antenna - wise rotati on Fig. 7. Illustration of tw o types of rotations in RA arrays. (a) Array- wise rotation. (b) Joint antenna-wise and array-wise rotation. One-Dimensional (1D) Antenna Rotation: In special cases where the an tenna performs only 1D rotation, the p oin ting and reference vectors are simplied to the follo wing forms. • Rotation around the x -axis shown in Fig. 6(b): θ n = ψ n = 0 , whic h gives the p oin ting vector  f ⊥ ,n = [1 , 0 , 0] T and the reference vector  f ∥ ,n = [0 , − s ϕ n , c ϕ n ] T . • Rotation around the y -axis shown in Fig. 6 (c): ϕ n = ψ n = 0 , which gives the p ointing v ector  f ⊥ ,n = [ c θ n , 0 , − s θ n ] T and the reference vector  f ∥ ,n = [ s θ n , 0 , c θ n ] T . • Rotation around the z -axis shown in Fig. 6(d): ϕ n = θ n = 0 , which giv es the p ointing vector  f ⊥ ,n = [ c ψ n , s ψ n , 0] T and the reference v ector  f ∥ ,n = [0 , 0 , 1] T . 2) Array Rotation: As illustrated in Fig. 7 , rotating the entire array not only c hanges the orientation of the an tennas but also alters their spatial positions relativ e to the arra y center. Let θ array ≜ [ ϕ array , θ array , ψ array ] T denote the rotation angle v ector of the array at the BS and let q B ,n ∈ R 3 × 1 denote the initial p osition of RA n . A ccordingly , the p osition of RA n after array rotation is giv en by ˜ q B ,n = R ( θ array )  q B ,n − q B , 0  + q B , 0 , ∀ n ∈ N , (6) where q B , 0 ∈ R 3 × 1 denotes the p osition of the array rotation cen ter at the BS. If all an tennas share identical initial orientations and no indep endent antenna-wise rotation is applied during arra y rotation as shown in Fig. 7 (a), then w e ha ve θ n = 0 , ∀ n ∈ N . In this case, all antennas still hav e the same p oin ting and reference v ectors after array rotation, which can be resp ectiv ely expressed as  f ⊥ ,n = R ( θ array )  f (0) ⊥ ,n = R ( θ array ) e 1 , ∀ n ∈ N , (7)  f ∥ ,n = R ( θ array )  f (0) ∥ ,n = R ( θ array ) e 3 , ∀ n ∈ N . (8) Ho wev er, if indep enden t antenna-wise rotation is p er- formed sim ultaneously during arra y rotation as sho wn in Fig. 7 (b), the rotation of b oth the arra y and an tenna elemen ts needs to b e considered, and the p ointing and reference v ectors of RA n can be expressed as  f ⊥ ,n = R ( θ array ) R ( θ n )  f (0) ⊥ ,n = R ( θ array ) R ( θ n ) e 1 , (9)  f ∥ ,n = R ( θ array ) R ( θ n )  f (0) ∥ ,n = R ( θ array ) R ( θ n ) e 3 , (10) resp ectiv ely . Based on the general 3D array rotation dened in ( 7 )–(10), the p ointing and reference vectors for sp ecic 2D and 1D array rotation special cases can also b e derived in a similar wa y as discussed for individual an tenna rotation. B. Channel Mo del Based on the antenna rotation mo del in Section I I-A, w e presen t the channel mo del for RA systems in this subsection. W e rst introduce t wo commonly used antenna directional gain patterns that c haracterize the radiation energy distribution. W e then construct the general near- eld line-of-sight (LoS) channel mo del for narrowband systems by incorp orating the RA directional gain, and pro vide its far-eld approximation. Finally , we extend the mo del to m ultipath propagation, wideband transmission, and p olarization eects, follo wed by a brief discussion of p oten tial extensions to other c hannel mo dels. 1) An tenna Directional Gain Pattern: Given the ori- en tation of eac h RA, the eective antenna gain dep ends on b oth the signal arriv al/departure direction and the underlying directional gain pattern. W e consider t wo commonly used mo dels: • Cosine pattern mo del: F or an antenna with a single narro w mainlob e and negligible side lob es, the direc- tional gain pattern can b e mo deled as [34] G ( ϵ, φ ) = ( G cos max cos 2 ρ ( ϵ ) , ϵ ∈ [0 , π 2 ) , φ ∈ [0 , 2 π ) 0 , otherwise , (11) where ( ϵ, φ ) is a pair of inciden t/departure angles of the signal with resp ect to the antenna’s current b oresigh t direction as illustrated in Fig. 8 (a), the parameter ρ ≥ 0 reects the antenna directivit y and mainlob e b eamwidth, and is determined by the radi- ation c haracteristic of the adopted antenna element via tting to measured or full-w av e sim ulated data. In addition, G cos max is the maxim um gain in the an tenna b oresigh t direction, and is given by G cos max = 2(2 ρ + 1) to meet the law of p o w er conserv ation. Moreov er, as sho wn in Fig. 8 (b), a larger ρ corresp onds to stronger directivit y , yielding higher antenna b oresigh t gain and a narrow er mainlobe. • 3GPP element mo del: Another practical and widely adopted mo del is the 3GPP directional gain pat- tern [74], [77], given by G ( ϵ, φ ) = G 3GPP max − min {− ( G e,H ( φ ) + G e,V ( ϵ )) , A max } , (12) 10 , k n u  , k n ϕ n f  y  n w B or e s i ght : , k n  User d i r ect i o n : x  z  ϕ  x  y  z  Bore s i ght di re c t i on S i gna l i nc i de nt / de pa rt ure di re c t i on (a) Inciden t/Departure angles for directional gain pattern 0 /4 /2 3 /4 5 /4 3 /2 7 /4 2 4 6 10 18 =0 =1/2 =1 =2 =4 (b) Directional gain patterns for dif- ferent parameters ρ Fig. 8. Illustration of antenna directional gain pattern. where G 3GPP max is the maximum gain and A max is the fron t-back attenuation limit. The horizontal and v ertical comp onen ts are resp ectively given by G e,H ( φ ) = − min ( 12  φ φ 3dB  2 , A max ) , (13a) G e,V ( ϵ ) = − min ( 12  ϵ ϵ 3dB  2 , A side ) , (13b) where φ 3dB and ϵ 3dB denote the 3 -dB b eam widths (t ypically 65°), and A side sp ecies the v ertical sidelob e suppression lev el. F or the cosine pattern mo del in (11), which is rotationally symmetric, the directional gain depends only on the angular oset ϵ betw een the signal direction and the an tenna b oresigh t. By con trast, for the 3GPP element mo del in (12), since the pattern is generally irregular, the directional gain dep ends on the sp ecic signal direction relativ e to the antenna b oresight, and thus b oth ϵ and φ need to b e taken into account. Nevertheless, b oth directional gain patterns ab ov e characterize the direction- dep enden t magnitude v ariation of the electric eld, while the phase in instan taneous channel mo deling can be absorb ed into the propagation c hannel co ecien t. 2) Near-Field LoS Channel Model: According to the directional gain patterns in (11) and (12), the eec- tiv e antenna gain at a spatial p oin t dep ends on the inciden t/departure angle pair ( ϵ, φ ) dened with resp ect to the an tenna’s curren t orientation. Let q U ,k ∈ R 3 × 1 denote the p osition of user k and  q U ,n,k ≜ q U ,k − q B ,n ∥ q U ,k − q B ,n ∥ denote the direction v ector from RA n to user k with k ∈ K ≜ { 1 , 2 , . . . , K } . Moreo v er, w e dene the orien tation v ector of RA n as  f n ≜ h  f T ⊥ ,n ,  f T ∥ ,n i T ∈ R 6 × 1 b y con- catenating the p oin ting and reference vectors, which can uniquely determine the an tenna orien tation. A ccordingly , the incident/departure angles of user k ’s propagation direction with resp ect to RA n are deriv ed as ϵ (  f n ,  q U ,n,k ) = arccos   q T U ,n,k  f ⊥ ,n  , (14a) φ (  f n ,  q U ,n,k ) = arctan2   q T U ,n,k  f ∥ ,n ,  q T U ,n,k ˜ e 2  , (14b) where ˜ e 2 ≜ R ( θ n ) e 2 denotes the rotated ˜ y -axis with e 2 ≜ [0 , 1 , 0] T . Therefore, the directional gain from RA n to user k can b e mo deled as g k   f n  = G  ϵ (  f n ,  q U ,n,k ) , φ (  f n ,  q U ,n,k )  . (15) This expression highlights that the antenna rotation (i.e., b y changing  f ⊥ ,n and/or  f ∥ ,n ) directly controls the eectiv e directional gain. F or analytical conv enience, the cosine pattern mo del (11), which is rotationally symmetric around the b oresigh t, is widely adopted [23], [24]. Under this practically useful mo del, (15) simplies to g k   f n  = G cos max h  q T U ,n,k  f ⊥ ,n i 2 ρ + , (16) where [ x ] + ≜ max { x, 0 } . It can b e inferred that the directional gain in (16) increases as the angle betw een the p oin ting vector  f ⊥ ,n and the user direction  q U ,n,k decreases. Intuitiv ely , the maximum directional gain G cos max can b e achiev ed when the RA boresight direction is aligned with the user direction, i.e.,  f ⊥ ,n =  q U ,n,k . A ccordingly , the near-eld LoS channel betw een RA n and user k is expressed as h LoS ,k   f n  = √ β 0 d n,k r g k   f n  e − j 2 π λ d n,k , (17) where β 0 ≜  λ 4 π  2 denotes the channel p o wer gain at a reference distance of d 0 = 1 meter (m) with λ b eing the carrier w av elength, d n,k ≜ k q U ,k − q B ,n k is the distance b et ween RA n and user k , and β n,k ≜ √ β 0 d n,k represen ts the propagation co ecien t. By stac king the c hannel coecients of all RAs in the array , the near-eld LoS channel vector b etw een the BS and user k , denoted b y h LoS ,k ( F ) ∈ C N × 1 , is given by h LoS ,k ( F ) = h h LoS ,k (  f 1 ) , h LoS ,k (  f 2 ) , . . . , h LoS ,k (  f N ) i T , (18) where F ≜ h  f 1 ,  f 2 , . . . ,  f N i ∈ R 6 × N denotes the stac ked orien tation matrix for all RAs. 3) F ar-Field LoS Channel Mo del: In man y practi- cal scenarios, since the arra y ap erture is m uch smaller than the link distance, the impinging wa vefron ts can be appro ximated as uniform plane w av es. Under this far- eld condition, the direction vectors and propagation co ecien ts across the RAs satisfy  q U , 1 ,k ≈  q U , 2 ,k ≈ · · · ≈  q U ,N ,k ≜  q U ,k and β 1 ,k ≈ β 2 ,k ≈ · · · ≈ β N ,k ≜ β k . T aking RA 1 as the reference, the array resp onse vector is th us giv en by a k ( N ,  q U ,k ) = h 1 , e j 2 π λ ( q B , 2 − q B , 1 ) T  q U ,k , . . . , e j 2 π λ ( q B ,N − q B , 1 ) T  q U ,k i T . (19) F or a uniform planar array (UP A), the ab o ve resp onse v ector can be transformed in to a k ( N ,  q U ,k ) = a y,k ( N y ,  q U ,k ) ⊗ a z ,k ( N z ,  q U ,k ) , (20) 11 where N y and N z are the num b ers of RAs along the y - and z -axes, resp ectively , and a y,k ( N y ,  q U ,k ) = h 1 , e j 2 π ∆ d λ  q T U ,k e 2 , . . . , e j 2 π ∆ d λ ( N y − 1)  q T U ,k e 2 i T , (21a) a z ,k ( N z ,  q U ,k ) = h 1 , e j 2 π ∆ d λ  q T U ,k e 3 , . . . , e j 2 π ∆ d λ ( N z − 1)  q T U ,k e 3 i T , (21b) are the 1D steering vector functions for uniform linear arra ys (ULAs) along the y - and z -axes, resp ectively , with ∆ d denoting the antenna spacing. Under the far-eld condition, the corresp onding direc- tional gain vector is g k ( F ,  q U ,k ) =         G  ϵ (  f 1 ,  q U ,k ) , φ (  f 1 ,  q U ,k )  G  ϵ (  f 2 ,  q U ,k ) , φ (  f 2 ,  q U ,k )  . . . G  ϵ (  f N ,  q U ,k ) , φ (  f N ,  q U ,k )          ∈ R N × 1 . (22) In particular, the directional gain vector under the cosine pattern model in (11) is given by g k ( F ,  q U ,k ) = G cos max          h  q T U ,k  f ⊥ , 1 i 2 ρ + h  q T U ,k  f ⊥ , 2 i 2 ρ + . . . h  q T U ,k  f ⊥ ,N i 2 ρ +          ∈ R N × 1 . (23) A ccordingly , the far-eld LoS channel vector b etw een the BS and user k can b e expressed as h k ( F ) = β k e − j 2 π λ d 1 ,k (diag( g k ( F ,  q U ,k ))) 1 2 a k ( N ,  q U ,k ) . (24) It can b e observed that the user direction aects b oth the directional gain and array resp onse v ectors. 4) Multipath Channel Mo del: F or scattering environ- men ts, w e adopt a geometric propagation mo del to charac- terize the multipath channel. W e assume that Q scatterer clusters are distributed in the propagation en vironment, where the position of cluster q is represented by q C ,q ∈ R 3 × 1 with q ∈ Q ≜ { 1 , 2 , . . . , Q } . Similar to (15), the directional gain from RA n to cluster q is expressed as ˜ g q   f n  = G  ϵ (  f n ,  q C ,n,q ) , φ (  f n ,  q C ,n,q )  , (25) where  q C ,n,q ≜ q C ,q − q B ,n ∥ q C ,q − q B ,n ∥ is the direction vector from RA n to scatterer cluster q . Under the cosine pattern mo del in (11), the directional gain from RA n to cluster q reduces to ˜ g q   f n  = G cos max h  q T C ,n,q  f ⊥ ,n i 2 ρ + . (26) Then, the non-line-of-sight (NLoS) channel b et ween RA n and user k is mo deled as h NLoS ,k   f n  = Q X q =1 σ q β 0 q ˜ g q (  f n ) ˜ d n,q ¯ d k,q e − j 2 π λ ( ˜ d n,q + ¯ d k,q ) , (27) where σ q denotes the radar cross section (RCS) of cluster q , and ˜ d n,q ≜ k q C ,q − q B ,n k and ¯ d k,q ≜ k q C ,q − q U ,k k denote the distances from RA n to scatterer cluster q and from scatterer cluster q to user k , resp ectiv ely . Thus, b y sup erimposing the LoS and NLoS comp onen ts, the ov erall m ultipath channel b et ween the BS and user k is given by h MP ,k ( F ) = h LoS ,k ( F ) + h NLoS ,k ( F ) ∈ C N × 1 , (28) with h NLoS ,k ( F ) = h h NLoS ,k (  f 1 ) , h NLoS ,k (  f 2 ) , . . . , h NLoS ,k (  f N ) i T (29) b eing the NLoS c hannel v ector. 5) Wideband Channel Model: F or RA-enabled wide- band communications, to eectively exploit the large bandwidth resources and ov ercome frequency-selective fading, w e consider an orthogonal frequency division m ultiplexing (OFDM) system with bandwidth B and L sub carriers. Let τ n,k, 0 = d n,k c and τ n,k,q = ˜ d n,q + ¯ d k,q c denote the propagation delay of the “RA n -user k ” and “RA n - cluster q -user k ” links, resp ectively , with c b eing the speed of light. A ccordingly , the spatial-time baseband equiv alent c hannel impulse resp onse betw een RA n and user k is giv en b y h WB ,k (  f n , t ) = Q X q =0 Γ k,q (  f n ) e − j 2 πf c τ n,k,q δ ( t − τ n,k,q ) , (30) where Γ k, 0 (  f n ) = β n,k q g k (  f n ) is the channel gain of the “RA n -user k ” link, Γ k,q (  f n ) = σ q β 0  ˜ d n,q ¯ d k,q  − 1 q ˜ g q (  f n ) with q = 1 , 2 , . . . , Q is the channel gain of the “RA n - cluster q -user k ” link, f c is the carrier frequency , and δ ( t ) denotes the Dirac delta function. The contin uous- time F ourier transformation (CTFT) of h WB ,k (  f n , t ) is then obtained as H WB ,k (  f n , f ) = Q X q =0 Γ k,q (  f n ) e − j 2 πf c ( 1+ f f c ) τ n,k,q , (31) whic h represents the spatial-frequency channel resp onse b et w een RA n and user k . Let ∆ f ≜ B L denote the OFDM sub carrier spacing. Then, the frequency of the l -th subcarrier is giv en by f l = ( l − 1)∆ f with l ∈ L ≜ { 1 , 2 , . . . , L } . Therefore, the spatial-frequency channel resp onse b etw een the BS and user k at sub carrier l is giv en b y h WB ,k,l ( F ) = h H WB ,k (  f 1 ,f l ) , H WB ,k (  f 2 ,f l ) , . . . , H WB ,k (  f N ,f l ) i T . (32) 6) Extension to Incorp orate P olarization Eects: Po- larization is a fundamen tal property of electromagnetic (EM) wa ves that describes the orien tation of the electric eld v ector during propagation. In wireless communica- tion systems, prop er p olarization alignment b etw een the transmit and receive antennas is essen tial for ecient 12 0 y 0 x 0 z U, , nk q  , n f  t , nk p P ol a ri z a t i on di re c t i on Ele c tr ic f ie ld d ir e c tio n P ropa ga t i on di re c t i on T a nge nt pl a ne E l ect r o m ag n et i c w av e Fig. 9. Illustration of p olarization directions of antenna and electric eld. t , 1 =0 0 nk      p r 1 =0 0 k      p ( ) ( ) t , 1 =0 0 nk      pR θθ ( ) t r , 1 T nk k ω = = pp P er f ect l y m at ch i n g ( ) ( ) tr , 1 T nk k ω ≤ = pp θ P o l a r i z a t i o n m i s m a t c h c a us e d by r ot a t i on Fig. 10. Illustration of rotation-induced p olarization mismatch. signal transmission and reception. In RA systems, antenna rotation c hanges the polarizat ion direction, which ma y impro ve or degrade the alignment b et ween the transmit and receive p olarizations, thereby aecting the eective c hannel gain [23], [78], [79]. T o incorporate p olarization eects into the RA channel mo del, we assume that b oth the BS and users employ linearly p olarized antennas. Additionally , the main p olar- ization direction of each RA is aligned with its reference v ector  f ∥ ,n . As illustrated in Fig. 9 , the electric eld at an observ ation p oin t is alwa ys orthogonal to the propagation direction and lies in the plane spanned b y the transmit p o- larization direction and the propagation direction. Thus, the eective transmit electric-eld vector, obtained b y pro jecting  f ∥ ,n on to the tangen t plane orthogonal to the propagation direction  q U ,n,k , is given by [23] p t n,k =  f ∥ ,n − (  f T ∥ ,n  q U ,n,k )  q U ,n,k . (33) This expression shows that the eective electric-eld direction dep ends jointly on the antenna p olarization direction and the user direction. As illustrated in Fig. 10, p erfect p olarization matching o ccurs when the electric-eld direction of the incom- ing wa ve aligns with the receive p olarization direction. Otherwise, p olarization mismatch reduces the eective receiv ed p o wer. Let p r k denote the receiv e p olarization direction at user k . The polarization matc hing eciency is characterized b y the inner pro duct b et w een p t n,k and p r k , yielding the p olarization matching gain [23] ω   f n ,  q U ,n,k  =  p t n,k  T p r k . (34) Then, b y incorp orating the polarization matching gain in to the channel mo del, the LoS channel co ecien t b e- t ween RA n and user k is giv en by h PE ,k   f n  = ω   f n ,  q U ,n,k  h LoS ,k   f n  . (35) This p olarization-aw are mo del shows that antenna rota- tion aects the c hannel amplitude through both direc- tional gain and p olarization matching. Therefore, prop er an tenna orientation control can balance these t w o eects to impro ve channel quality . The ab o ve p olarization-a ware channel mo del applies to linearly polarized an tennas with arbitrary slan t angles, including purely v ertical or horizontal polarization. It can also b e extended to cross-p olarized (dual-p olarized) an tenna pairs, such as ± 45 ◦ p olarized elements. In suc h cases, each element can b e treated as a linearly p olarized an tenna sharing the same RF c hain, and its c hannel can be mo deled using (35). By introducing p olarforming technol- ogy that jointly controls the t wo elements, the com bined transmit/receiv e p olarization can b e adaptively adjusted through phase and amplitude control [80]. F urthermore, although (35) is derived for narrowband LoS channels, it can b e extended to multipath and wideband scenarios by incorp orating the p olarization eects of scatterer clusters in to the multipath mo del in (28) and the wideband mo del in (30). 7) Extensions to Other Channel Mo dels: The RA c hannel models developed ab ov e follo w a geometry-/eld- resp onse formulation, where the channel explicitly de- p ends on the an tenna orientations in a deterministic prop- agation environmen t. This enables direct optimization of RA orientations to enhance comm unication and sensing p erformance. Additionally , the baseline mo del assumes an RA array deplo yed at the BS and a single xed antenna at each user. It can b e readily generalized to MIMO links where users are also equipp ed with RA arra ys. In such cases, the MIMO channel can be constructed b y stac king the SISO channels corresp onding to each transmit–receiv e RA pair, and the directional gain must accoun t for rotations at b oth ends. Bey ond deterministic environmen ts, the RA channel mo del can b e extended to stochastic channel models by treating user locations, scatterer positions, and small-scale fading co ecien ts as random v ariables. F or example, under a Rician fading mo del, the c hannel can b e decomp osed in to a deterministic LoS comp onent mo deled by (17) and a sto c hastic NLoS comp onen t capturing spatially correlated m ultipath. These statistics are often assumed stationary or quasi-static, but in practice, mobilit y of the transceiv ers or surrounding ob jects induces spatio-temp oral v ariations and Doppler shifts. Suc h dynamics can be incorp orated b y mo deling the time-v arying geometry and scheduling RA orien tations to trac k c hannel ev olution. Finally , the RA mo deling framework naturally extends 13 to distributed antenna deploymen ts, suc h as cell-free MIMO systems [81], [82]. In this case, the ov erall c hannel is formed by aggregating orientation-dependent channel resp onses across geographically distributed access p oin ts (APs), where coordinated orien tation control inuences spatial correlation, macro-diversit y , and cov erage perfor- mance. C. Optimization F ramew ork and Design Issues T o fully exploit the additional spatial DoF s in tro duced b y antenna rotation, we consider a generic optimization framew ork for RA-enabled wireless systems: max Θ , S U ( Θ , S ) (36a) s.t. F i ( Θ ) ≥ 0 , 1 ≤ i ≤ I F , (36b) G i ( S ) ≥ 0 , 1 ≤ i ≤ I G , (36c) Q i ( Θ , S ) ≥ 0 , 1 ≤ i ≤ I Q , (36d) where Θ ≜ [ θ 1 , θ 2 , . . . , θ N , θ array ] ∈ R 3 × ( N +1) collects the rotation angles of all RAs and the arra y platform, and S denotes the set of system resources (e.g., transmit pow er, bandwidth, b eamforming v ectors, user asso ciation). The n umber of constrain ts I X with X ∈ { F , G , Q } dep ends on the system conguration, including the num b ers of RAs and users. The utility function U ( · ) quan ties system p er- formance; F i ( · ) captures an tenna/array rotation-related constrain ts (e.g., limited rotation range, discrete antenna rotation, minimum antenna spacing, and maximum an- tenna rotation sp eed); G i ( · ) captures resource limitations; and Q i ( · ) represents coupled constraints inv olving b oth rotation and resource allo cation. 1) Utility F unction: The utilit y function U ( Θ , S ) in (36a) reects the p erformance ob jective of the RA-enabled system. In communication systems, it may represent the ac hiev able rate, secrecy rate, co verage probability , or outage p erformance [24], [25], [ 81], [ 83]. In sensing systems, t ypical metrics include the Cramér–Rao bound (CRB), detection probability , and estimation accuracy [84], [85]. System-lev el metrics suc h as energy eciency , sp ectral eciency , or latency can also b e incorp orated dep ending on the net work architecture and service requiremen ts [79], [86], [87]. F or ISA C systems, U ( · ) ma y be designed to strik e a balance b etw een comm unication throughput and sensing accuracy [88], [89]. In intelligen t computing netw orks, it may further include computation latency , computing eciency , and task completion rate, thereby capturing the in terplay betw een communication and computation resources [90]. 2) Antenna/Arra y Rotation Constraint: Regardless of whether mec hanical or electronic rotation is used, the feasible rotation range is physically limited. T o ensure that each RA or the array platform op erates within the resp ectiv e hardware constraints, the rotation angles m ust satisfy [ θ low er ,n ] i ≤ [ θ n ] i ≤ [ θ upper ,n ] i , ∀ n ∈ N , i ∈ { 1 , 2 , 3 } , (37a) [ θ low er , array ] i ≤ [ θ array ] i ≤ [ θ upper , array ] i , i ∈ { 1 , 2 , 3 } , (37b) where θ low er ,n (or θ low er , array ) and θ upper ,n (or θ upper , array ) denote the lo wer and upp er b ounds of the rotation angles of the RA n /array , resp ectively . In compact arra ys, large an tenna rotations ma y increase m utual coupling and distort radiation patterns due to spatial ov erlap. T o a void excessive b oresigh t deviation and mitigate coupling eects, an additional constraint is imp osed [23], [24]: 0 ≤ arccos   f T ⊥ ,n e 1  ≤ θ max , ∀ n ∈ N , (38) where θ max ∈ [0 , π 2 ] sp ecies the maximum allo wable deviation of each RA’s boresight from its initial nominal direction (i.e., the p ositiv e x -axis). 3) Other Constraints: Constraints G i ( S ) ≥ 0 represen t system resource limitations, such as maximum transmit p o w er, av ailable bandwidth, b eamforming constrain ts, and maximum transmission duration. These constraints c haracterize the physical and op erational limitations of the RA system and ensure that resource usage remains within practical and regulatory b ounds. In addition, constrain ts Q i ( Θ , S ) ≥ 0 capture the joint impact of an tenna orientation and resource allo cation. Examples include minimum signal-to-in terference-plus-noise ratio (SINR) requirements, minim um received signal p ow er, maxim um tolerable sensing error, and outage probability constrain ts. These constraints reect quality-of-service (QoS) and reliability requirements for dynamic an tenna rotations. Compared with xed-antenna systems, RA systems can signicantly enhance p erformance by reconguring directional gain patterns and concentrating radiation en- ergy tow ard desired directions. Ho wev er, antenna/arra y rotation in tro duces additional implementation cost, en- ergy consumption, latency , and computational complexity . Practical RA optimization must therefore account for hardw are imp erfections (e.g., mec hanical inaccuracies, RF distortions), imp erfect channel state information (CSI) due to estimation errors and channel v ariations, and nite rotation sp eed [25]. These factors inuence the ac hiev able p erformance and motiv ate robust optimization framew orks that ensure reliable operation under real-world constrain ts. I II. Antenna/Arra y Rotation Optimization Determining the optimal antenna orien ta- tions/b oresigh ts is essential for fully exploiting the p erformance gains oered by RA systems. Since wireless c hannels dep end on antenna/arra y orientation in a highly nonlinear manner, the resulting optimization problems are often c hallenging and require carefully designed solution strategies. In this section, w e introduce ecien t optimization methods for antenna/arra y rotation in v arious wireless system settings and demonstrate their p erformance adv antages ov er conv entional xed- an tenna arc hitectures. T o c haracterize the fundamen tal p erformance b ounds of RA systems, we assume p erfect CSI throughout this section. Practical issues related to 14 RA c hannel estimation and acquisition will b e addressed in Section IV. A. RA-Enabled MISO/SIMO System In MISO/SIMO systems, deploying an RA array at the BS enables signicant array-gain enhancemen t b y join tly adjusting the orientations/boresights of all anten- nas. The resulting spatial DoF s allow the RA array to concen trate radiated energy to ward the intended user, thereb y reducing energy leakage and improving trans- mission eciency . F or ease of exp osition, we focus on a single-user MISO system ( K = 1 ) under free-space and narro wband propagation. W e adopt the cosine pattern mo del in (11) and assume that the user employs an isotropic antenna. Owing to uplink–do wnlink duality , the analytical framew ork directly extends to the SIMO case with receiv e beamforming. Under the cosine pattern mo del, the directional gain dep ends solely on the pro jection betw een the p oin ting v ector and the user direction, as shown in (16). Hence, a 2D rotation mo del suces to characterize RA orientation, and optimizing the rotation angles { θ n } is equiv alent to optimizing the p oin ting vectors {  f ⊥ ,n } , thereby mitigating the coupling among the rotation angles. Dropping the user index in the near-eld LoS channel mo del (18) for simplicit y , the signal-to-noise ratio (SNR) maximization problem is formulated as follows: (P-MISO): max w , {  f ⊥ ,n } γ = 1 σ 2   w T t h LoS ( F )   2 (39a) s.t. 0 ≤ arccos (  f T ⊥ ,n e 1 ) ≤ θ max , ∀ n, (39b) k  f ⊥ ,n k = 1 , ∀ n, (39c) k w t k 2 ≤ P , (39d) where w t ∈ C N × 1 is the transmit b eamforming v ector, and P and σ 2 denote the maximum transmit p o wer and the noise p ow er, resp ectiv ely . In addition, constraint (39c) ensures that  f ⊥ ,n is a unit-norm vector, and constraint (39d) guarantees that the transmit p ow er do es not exceed its budget. F or an y giv en RA p ointing vectors {  f ⊥ ,n } , the opti- mal transmit b eamformer for problem (P-MISO) can b e obtained via maximum-ratio transmission (MR T), i.e., w MR T = √ P h ∗ LoS ( F ) ∥ h LoS ( F ) ∥ . Thus, substituting w MR T in to (39a) yields γ = P σ 2 k h LoS ( F ) k 2 = P β 0 G cos max σ 2 N X n =1 1 d 2 n h  q T U ,n  f ⊥ ,n i 2 ρ + . (40) Since eac h term in (40) dep ends only on the pointing v ector of RA n , problem (P-MISO) can b e decomposed in to N indep enden t subproblems: max  f ⊥ ,n  q T U ,n  f ⊥ ,n (41a) s.t. 0 ≤ arccos(  f T ⊥ ,n e 1 ) ≤ θ max , (41b) k  f ⊥ ,n k = 1 , (41c) ( 0, 0, ) d n   ma x θ s p an ( ) N ∆    O x y ma x θ R A o r ien tatio n /b o r es ig h t Us er d ir ectio n d N ∆ Fig. 11. Illustration of the geometric relationship b et ween the user and RAs in the considered RA-enabled MISO system. where the constan t term is omitted in (41a). Maximizing the pro jection b etw een  f ⊥ ,n and  q U ,n yields the following optimal pointing vector [23], [24]:  f ⋆ ⊥ ,n =  cos θ ⋆ z ,n , sin θ ⋆ z ,n cos θ ⋆ a ,n , sin θ ⋆ z ,n sin θ ⋆ a ,n  T , (42) where θ ⋆ z ,n and θ ⋆ a ,n are giv en b y θ ⋆ z ,n = min  arccos   q T U ,n e 1  , θ max  , (43a) θ ⋆ a ,n = arctan2   q T U ,n e 3 ,  q T U ,n e 2  , (43b) resp ectiv ely . This indicates that each RA should align its b oresigh t with the user direction as closely as allow ed by the rotation constraints to maximize its directional gain. F or a ULA-based MISO system with antenna directivit y factor ρ = 1 2 and a user lo cated directly in front of the arra y as sho wn in Fig. 11, the analysis of [23] sho w that the maxim um SNR based on the optimal RA pointing v ector in (42) admits the follo wing closed-form expression: γ = ( 2 ζ P π 2 σ 2 4 span ( N ) , N ≤ ¯ N 2 ζ P π 2 σ 2 [ θ max + sin ( 4 span ( N ) − θ max )] , N > ¯ N , (44) where ζ ≜ ∆ d d  1 with d b eing the distance b et ween the cen ter of the ULA and the user, 4 span ( N ) ≜ arctan  N ζ 2  denotes the user span angle, dened as the angle b etw een the t wo line segments connecting the user to the cen ter and to one end of the ULA as illustrated in Fig. 11, and ¯ N ≜ 2 j tan θ max ζ k + 1 is the maximum num b er of an tennas whose b oresigh ts can b e aligned with the user direction. The maximum SNR in (44) scales with the num b er of RAs N according to the span angle 4 span ( N ) and is fundamen tally limited by the allow able rotation range. F urthermore, as N → ∞ , the asymptotic SNR b ecomes lim N →∞ γ = 2 ζ P π 2 σ 2 ( θ max + cos θ max ) . (45) It is observ ed that a larger rotational range yields a higher asymptotic SNR. SNR Scaling With N : Using the linear appro ximation for the arctangent function, i.e., arctan( x ) ≈ π 4 x, − 1 ≤ x ≤ 1 [91], the SNR in the rst case of (44) can be approx- imated b y γ ≈ P ζ 2 4 πσ 2 N since we hav e 0 ≤ N ζ 2 ≤ 1 when N ≤ ¯ N . Thus, the SNR grows linearly with the num b er of RAs as long as N ≤ ¯ N , i.e., 4 span ( N ) ≤ θ max . In addition, it can b e v eried that f ( x ) ≜ sin (arctan ( x ) − θ max ) is a 15 10 2 10 3 -40 -35 -30 -25 -20 -15 5 dB 1.5 dB Fig. 12. Received signal p o wer versus the n umber of antennas N for the RA-enabled MISO system. conca ve increasing function, and lim x →∞ f ′ ( x ) = 0 . This indicates that when N > ¯ N , i.e., 4 span ( N ) > θ max , the gro wth rate of the maximum SNR gradually decreases as the num b er of RAs increases, even tually approaching zero. As sho wn in [23], a similar SNR scaling law can b e obtained for the general UP A case. T o ev aluate the arra y-gain improv ement ac hieved b y RAs, w e consider a single-user MISO system with a ULA transmitter and a user lo cated d = 15 m from the array cen ter. The system op erates at 2.4 GHz ( λ = 0 . 125 m), with noise pow er σ 2 = − 80 dBm, an tenna spacing ∆ d = λ/ 2 , transmit p o wer P = 10 dBm, and maximum deviation angle θ max = π / 6 . F or these settings, Fig. 12 sho ws the received signal p o wer v ersus the num b er of an tennas N for b oth RA-enabled and xed-an tenna MISO systems. F or small to mo derate N , both the RA and xed- an tenna systems exhibit linear p ow er scaling. Ho wev er, as N increases further, the received signal p o wers of b oth systems even tually approach their asymptotic v alues, whic h v alidates the SNR scaling law discussed ab o ve. A dditionally , the RA system consisten tly outp erforms the xed-an tenna system b ecause each RA can indep enden tly steer its boresight tow ard the user to ac hieve a higher directional gain. The p erformance adv antage of the RA system ov er its xed-antenna counterpart is particularly pronounced (up to 5 dB) when N ≤ 100 . This indicates that, by more eectiv ely concen trating radiated energy to ward the user lo cation, the RA architecture can achiev e a higher SNR with fewer antennas, esp ecially when the arra y size is constrained. F urthermore, when N → ∞ , the RA system attains an asymptotic gain of nearly 1.5 dB o ver the xed-antenna system. This underscores the fact that the inherent adv antages of RA are not ov ershadow ed ev en with a larger num b er of antennas. B. RA-Enabled MIMO System Compared with MISO/SIMO systems where an- tenna/arra y rotation is emplo yed only at one side of the link, RA-enabled MIMO architectures allo w join t orien tation/b oresigh t adjustment at b oth transmitter and receiv er. The resulting additional spatial DoF s enable dy- namic b eam alignmen t with dominan t propagation paths, mitigating angular mismatch caused b y user mobility or environmen tal v ariations. By jointly optimizing the transmit- and receiv e-side RA orien tations/b oresigh ts, the system can sim ultaneously concentrate radiated energy to ward desired directions and enhance the receiv ed signal p o w er at the target user. As a result, the ov erall c hannel condition improv es, inter-path correlation is reduced, and the MIMO c hannel matrix b ecomes higher-rank and better conditioned. This transceiver-side spatial reconguration strengthens the spatial m ultiplexing capability and ulti- mately increases the achiev able channel capacity . T o illustrate these adv antages, we consider an RA- enabled MIMO comm unication system, where transmit- ter and receiver are equipp ed with N t and N r RAs, resp ectiv ely . Let H ( F t , F r ) ∈ C N r × N t denote the MIMO c hannel matrix, where F t ∈ R 6 × N t and F r ∈ R 6 × N r are the orien tation matrices of the transmit and receive RAs, resp ectiv ely . In particular, the ( n r , n t ) -th entry of H ( F t , F r ) corresp onds to the channel coecient betw een the n t -th transmit RA and the n r -th receive RA, where n t ∈ N t ≜ { 1 , 2 , . . . , N t } and n r ∈ N r ≜ { 1 , 2 , . . . , N r } . The c hannel co ecient of each transmit-receive RA pair can b e mo deled using the LoS, m ultipath, wideband, and p olarization-a w are channel mo dels in (17), (27), (30), and (35), respectively . Giv en channel matrix H ( F t , F r ) and transmit cov ari- ance matrix S ∈ C N t × N t with S  0 , the MIMO c hannel capacit y is giv en b y C MIMO ( F t , F r , S ) = log 2 det  I N r + 1 σ 2 H ( F t , F r ) SH ( F t , F r ) H  . (46) Based on the cosine pattern mo del in (11), we assume that the initial b oresights of the transmit and receive RAs are aligned with the positive and negative x -axis, resp ectiv ely . The corresponding MIMO c hannel capacity maximization problem join tly optimizes the transmit and receiv e RA orientation matrices F t and F r , as well as the transmit co v ariance matrix S . Then, the resulting problem is form ulated as (P-MIMO): max F t , F r , S ⪰ 0 C MIMO ( F t , F r , S ) (47a) s.t. 0 ≤ arccos  (  f (t) ⊥ ,n t ) T e 1  ≤ θ max , ∀ n t ∈ N t , (47b) 0 ≤ arccos  − (  f (r) ⊥ ,n r ) T e 1  ≤ θ max , ∀ n r ∈ N r , (47c) k  f (t) ⊥ ,n t k = 1 , ∀ n t ∈ N t , (47d) k  f (r) ⊥ ,n r k = 1 , ∀ n r ∈ N r , (47e) T r( S ) ≤ P , (47f ) where constraint (47f) ensures that the transmit p o wer do es not exceed its maximum v alue P . Problem (P-MIMO) is challenging to solv e due to the highly non-concav e ob jective in (47a). A practical 16 approac h is to apply the blo ck co ordinate descent (BCD) tec hnique. Sp ecically , for xed RA orientation matrices F t and F r , the subproblem of optimizing the transmit co v ariance matrix S is a conv ex problem that can b e solv ed via Lagrange duality or the CVX solver. Ho wev er, for xed transmit co v ariance matrix S , the subproblem of optimizing the RA orientation matrices F t and F r remains non-con vex and dicult to solve. T o tackle this diculty , successiv e conv ex approximation (SCA) or gradient-based metho ds can be used to obtain high-qualit y suboptimal so- lutions for F t and F r . A dditionally , a high-qualit y solution for the transmitting and receiving orientation matrices can b e obtained by alternating optimization (AO) b etw een F t and F r in an iterative manner, which oers low er computational complexity . This iterative approach mirrors the practical deplo yment strategy of adjusting an tenna orien tations alternately at the transmitting and receiving ends. C. RA-Enabled Multi-User System The enhanced array and multiplexing gains enabled b y the RA architecture can also b e lev eraged in m ulti- user communication systems. In single-user scenarios, all RAs can orien t their b oresigh ts tow ard the same user to maximize the directional gain. In contrast, in multi-user settings, dierent RAs can steer their b oresigh ts in dif- feren t user directions, thereby enhancing directional gains for multiple users simultaneously . Moreov er, in multipath en vironments, RA orientations/boresights can b e adapted to the spatial distribution of the us ers and scatterers to balance m ultiple propagation paths and maximize the eectiv e channel gain. T o illustrate this capability , w e consider an RA-enabled m ulti-user uplink system op erating in a m ultipath channel, where K users (each equipp ed with a single isotropic an tenna) sim ultaneously transmit to a BS equipped with a UP A comprising N RAs. Using the multipath channel mo del in (28), the SINR for deco ding the signal of user k at the BS is giv en by γ k = P k | w T r ,k h MP ,k ( F ) | 2 P j  = k P j | w T r ,k h MP ,j ( F ) | 2 + σ 2 , (48) where P k is the transmit p ow er of user k , and w r ,k ∈ C N × 1 is the linear receive b eamforming vector for user k with k w r ,k k = 1 . A dopting the cosine pattern mo del in (11), we for- m ulate the following max–min SINR problem b y jointly optimizing the receive b eamforming matrix W r ≜ [ w r , 1 , w r , 2 , . . . , w r ,K ] ∈ C N × K and the RA p oin ting v ec- tors {  f ⊥ ,n } : (P-MU): max W r , {  f ⊥ ,n } min k γ k (49a) s.t. 0 ≤ arccos(  f T ⊥ ,n e 1 ) ≤ θ max , ∀ n, (49b) k  f ⊥ ,n k = 1 , ∀ n, (49c) k w r ,k k = 1 , ∀ k . (49d) 0 0.5 1 1.5 2 2.5 3 3.5 4 10.5 11 11.5 12 12.5 13 13.5 Fig. 13. Max-min achiev able rates of dierent systems versus the antenna directivity factor ρ . Since the optimization v ariables are tigh tly coupled and the ob jectiv e in (49a) is highly non-conv ex, it is di- cult to obtain the optimal solution of problem (P-MU) directly . T o address this issue, an AO algorithm can b e dev elop ed to alternately optimize the receiv e beamforming and the RA p oin ting vectors in an iterative manner. Sp ecically , for xed RA p oin ting vectors, linear receivers suc h as zero-forcing (ZF) and minim um mean-square error (MMSE) b eamforming can b e used to enhance the SINR of eac h user. On the other hand, for xed ZF/MMSE b eamforming, the RA p oin ting-v ector subproblem remains non-con vex. By relaxing (49c) to k  f ⊥ ,n k ≤ 1 , the SCA tec hnique can b e applied to obtain a locally optimal solution iteratively . F urthermore, for the sp ecial case of ρ = 1 in (16) and (26), problem (P-MU) can b e reform ulated as a conv ex semidenite program (SDP), whic h can b e solv ed eciently without iteration, as shown in [23]. T o ev aluate the p erformance gain of RAs in m ulti- user communication under multipath propagation, Fig. 13 depicts the max–min achiev able rate C MU = min k log 2 (1 + γ k ) = log 2 (1 + min k γ k ) , (50) for dierent schemes versus the an tenna directivit y factor ρ . Sp ecically , a square UP A with N = 4 × 4 RAs is equipp ed at the BS, serving K = 4 users uniformly distributed in four distinct directions in fron t of the BS. User distances are drawn uniformly from the interv al [30 , 50] m, and Q = 8 scatterer clusters are randomly placed around the users. The maximum deviation angle is θ max = π / 6 , and the transmit p o wer of each user is P k = 10 dBm. The results in Fig. 13 show that the max–min ac hiev able rate of the RA system increases with ρ , since a larger directivit y factor yields higher b oresigh t gain and a narro wer mainlob e. This enables the RA array to more eectiv ely achiev e directional gains across multiple user directions, thereby pro viding a higher max-min achiev able rate. In con trast, for xed-antenna systems, the max–min 17 rate decreases when ρ ≥ 1 , since a narro wer mainlob e reduces the directional gain for users located a wa y from the arra y’s main p oin ting direction. Additionally , although the random orien tation design can disperse the radiation p o w er in sev eral directions, it is signican tly inferior to the RA system since it fails to strategically allo cate antenna orien tations to balance directional gains across multipath c hannels and fairly impro ve the communication p erfor- mance of all users. These results highlight the imp ortance of RA arc hitectures for improving multi-user performance, esp ecially when antennas exhibit strong directivity and narro w mainlob es. D. RA-Enabled Wideband System By dynamically adjusting an tenna orienta- tions/b oresigh ts to capture the dominan t propagation paths while atten uating dela yed or w eaker multipath comp onen ts arriving from other directions, RAs provide additional spatial DoF s that can b e exploited to enhance o verall system performance across multiple frequency bands in wideband systems. This spatial selectivit y helps mitigate frequency-selectiv e fading and yields a more fa vorable eective channel resp onse across sub carriers. In this subsection, w e study an RA-enabled wideband OFDM system to demonstrate ho w RA architectures balance m ultiple propagation paths and enhance wideband comm unication p erformance. Extending the multi-user system to the wideband set- ting, we consider an OFDM channel with L sub carriers as mo deled in (32). Let a k,l b e a binary v ariable indicating whether subcarrier l with l ∈ L is assigned to user k ( a k,l = 1 ) or not ( a k,l = 0 ). If a k,l = 1 and p erfect syn- c hronization (including sym b ol timing, frame alignment, and carrier-frequency-oset comp ensation) is assumed at the BS, the receiv ed signal from user k on subcarrier l is giv en by ˜ y k,l = h WB ,k,l ( F ) p P k,l s k,l + n l , (51) where s k,l and P k,l denote the transmitted symbol and p o w er of user k on sub carrier l , resp ectiv ely , and n l ∼ N c ( 0 , ˜ σ 2 l I N ) is the additive white Gaussian noise (A W GN) v ector of subcarrier l with v ariance ˜ σ 2 l . Since each sub carrier is allo cated to at most one user, there is no inter-user in terference on any sub carrier. Th us, the BS can apply maxim um-ratio combining (MRC), i.e., w r ,k,l = h ∗ WB ,k,l ( F ) ∥ h WB ,k,l ( F ) ∥ , to maximize the receive SNR. The ac hiev able rate of user k in bits p er second p er Hertz (bps/Hz) is then given by ˜ R k = 1 L + L CP L X l =1 a k,l log 2 1 + P k,l k h WB ,k,l ( F ) k 2 ˜ σ 2 l ! , (52) where L CP is the cyclic-prex (CP) length. W e no w formulate the sum-rate maximization problem b y jointly optimizing the RA p oin ting vectors {  f ⊥ ,n } and 16 32 48 64 80 96 112 128 9 10 11 12 13 14 15 Fig. 14. Ac hievable sum-rates of dieren t schemes versus the num b er of the subcarriers L . the subcarrier allo cation v ariables { a k,l } . (P-WB): max {  f ⊥ ,n } , { a k,l } K X k =1 ˜ R k (53a) s.t. 0 ≤ arccos(  f T ⊥ ,n e 1 ) ≤ θ max , ∀ n, (53b) k  f ⊥ ,n k = 1 , ∀ n, (53c) K X k =1 a k,l ≤ 1 , ∀ l , (53d) L X l =1 a k,l P k,l ≤ P k , ∀ k , (53e) a k,l ∈ { 0 , 1 } , ∀ k , l , (53f ) where constraint (53d) ensures that each sub carrier is allo cated to at most one user to av oid inter-user interfer- ence, constraint (53e) guarantees that the total transmit p o w er of user k across all subcarriers do es not exceed its maxim um v alue P k , and constraint (53f) ensures that a k,l is a binary v ariable. Problem (P-WB) is non-conv ex due to the coupling b et w een the RA orientation and sub carrier allo cation. T o tac kle this diculty , we can alternately optimize the sub carrier allo cation { a k,l } and the RA p oin ting vectors {  f ⊥ ,n } in an iterative manner. On the one hand, for xed RA p ointing vectors {  f ⊥ ,n } , the subproblem of optimizing { a k,l } is a binary integer program that can b e solved using standard optimization to ols. On the other hand, for xed { a k,l } , the subproblem of optimizing {  f ⊥ ,n } remains highly non-con vex. By approximating the ob jective in (53a) as a concav e function with resp ect to {  f ⊥ ,n } and relaxing equality constrain t (53c), the SCA technique can b e applied to solv e this subproblem eciently . T o v alidate the b enets of the RA architecture in wideband systems, we consider an OFDM system with carrier frequency f c = 2 . 4 GHz, bandwidth B = 40 MHz, L = 64 sub carriers, and CP length L CP = 6 . Other system parameters follo w those of the previous m ulti-user setup. 18 Fig. 14 sho ws the achiev able sum-rate versus the num b er of sub carriers. It is observed that the sum-rate increases with L for all schemes because the relative CP ov erhead decreases as the num b er of sub carriers gro ws, improving sp ectral eciency . A dditionally , for RA systems, a larger L also provides greater exibility in sub carrier allo cation, whic h can b e jointly optimized with antenna orientations. The RA-enabled wideband system consistently achiev es the highest sum-rate b y join tly optimizing pointing vectors and sub carrier assignment. This p erformance gain stems from the abilit y of RAs to exibly adjust their b oresigh ts to balance m ultiple propagation paths across sub carriers, thereb y improving the o verall wideband c hannel quality . E. RA-Enabled ISAC System ISA C is widely regarded as a k ey enabling tec hnology for 6G wireless net works, oering unied communication and sensing functionalities through shared hardw are resources. Lev eraging the additional spatial DoF s introduced by an- tenna rotation, RA architectures can adaptly control their b oresigh ts according to ISAC p erformance requirements and the spatial distribution of communication users and sensing targets. Sp ecically , by steering antenna orien ta- tions/b oresigh ts tow ard in tended users or regions of inter- est, an RA array can signican tly enhance communication capacit y and link reliability , while also increasing the echo p o w er for sensing, thereby improving detection accuracy and sensing range. F urthermore, joint optimization of RA orien tations/b oresights and probing signals enables exible trade-os b et ween communication cov erage and sensing p erformance, improving b oth sp ectral and energy eciency [88]. T o examine the p oten tial of RAs for ISAC applications, w e consider an RA-enabled ISAC system in whic h a BS transmits b oth communication signals and dedicated probing signals to simultaneously serve K downlink users and sense a p oten tial target lo cated within a region A . The BS transmits and receives probing signals to estimate the target’s p osition. T o av oid self-interference b et ween comm unication and sensing, we assume that they share the same BS architecture but use separate time or frequency resources. Let q U ,k ∈ R 3 × 1 denote the p osition of user k . By denoting the transmit b eamforming v ector for user k as w t ,k ∈ C N × 1 , the ac hiev able rate of user k can b e expressed as ¯ R k ( F , W t ) = log 2 1 + | h T ( F , q U ,k ) w t ,k | 2 P j  = k | h T ( F , q U ,k ) w t ,j | 2 + σ 2 k ! , (54) where h ( F , q ) denotes the c hannel v ector betw een the RA- based BS and any spatial p oin t q ∈ R 3 × 1 , whic h can b e constructed using the c hannel mo dels in Section I I-B, W t ≜ [ w t , 1 , w t , 2 , . . . , w t ,K ] ∈ C N × K is the transmit b eamforming matrix, and σ 2 k is the noise p o wer at user k . F or sensing, we mo del the p oten tial target as an unstructured p oin t lo cated at q T ∈ A . The round-trip c hannel matrix for the BS to receive the probing signal reected b y the target, denoted by ¯ H ( F , q T ) ∈ C N × N , is expressed as ¯ H ( F , q T ) = σ T h ( F , q T ) h T ( F , q T ) , q T ∈ A , (55) where σ T denotes the R CS of the target. Let T s denote the total n umber of symbols during one sensing p erio d. Let s [ t ] ∈ C N × 1 represen t the t -th transmit probing symbol sen t b y the BS, where t ∈ { 1 , 2 , . . . , T s } . Using the ec ho p o w er as a practical sensing metric, the received ec ho p o w er at q T is giv en b y ¯ P ( F , ¯ S , q T ) = E  k ¯ H ( F , q T ) s [ t ] k 2  = | σ T | 2 k h ( F , q T ) k 2 h T ( F , q T ) ¯ Sh ∗ ( F , q T ) , (56) where ¯ S ≜ E  s [ t ] s H [ t ]  is the probing-signal co v ariance matrix. A dditionally , the CRB, whic h is widely adopted as a theoretical b ound for lo calization p erformance, can also serv e as a metric to c haracterize the sensing p erformance in the RA-enabled ISA C systems, as shown in [85]. In general, a higher echo p o wer received from the target leads to a low er CRB, and thus b etter sensing p erformance. T o ac hieve consisten t regional sensing o ver A while guaran teeing communication p erformance, w e jointly op- timize the RA orien tation matrix F , the communication b eamforming matrix W t , and the probing-signal cov ari- ance matrix ¯ S under the cosine pattern mo del in (11). The goal is to maximize the minimum received echo p o w er ov er A while ensuring that each user meets a minim um communication-rate requiremen t. The resulting optimization problem is formulated as ( P-ISA C ) : max F , W t , ¯ S ⪰ 0 min q T ∈A ¯ P ( F , ¯ S , q T ) (57a) s.t. ¯ R k ( F , W t ) ≥ R min , ∀ k , (57b) K X k =1 k w t ,k k 2 ≤ P max , c , (57c) T r( ¯ S ) ≤ P max , s , (57d) 0 ≤ arccos(  f T ⊥ ,n e 1 ) ≤ θ max , ∀ n, (57e) k  f ⊥ ,n k = 1 , ∀ n, (57f ) where constraint (57b) ensures that user k is served with its required minimum communication rate R min , and constrain ts (57c) and (57d) preven t the a verage transmit p o w ers of the communication and sensing signals from exceeding their budgets P max , c and P max , s , respectively . Problem (P-ISAC) is dicult to solve due to the non-con vexit y of ( 57a), (57b), and (57f), as well as the semi-innite nature of (57a) since A is contin uous. T o address this issue, we discretize A into M sampling p oin ts { q T ,m } M m =1 , yielding the approximate problem: ( P-ISA C-D ) : max η, F , W t , ¯ S ⪰ 0 η (58a) s.t. ¯ P ( F , ¯ S , q T ,m ) ≥ η , ∀ m, (58b) (57b) − (57f) , (58c) 19 2 3 4 5 6 7 8 9 10 0 0.5 1 1.5 2 2.5 3 3.5 10 -5 Fig. 15. Echo signal p ow er of dierent systems versus the required minimum communication rate R min . where η denotes the minim um received echo signal p ow er, and q T ,m with m ∈ M ≜ { 1 , 2 , . . . , M } is the p osition of the m -th spatial sampling p oint. Then, to address the non-con vexit y of problem (P-ISAC-D), we adopt a BCD framew ork, which alternately optimizes { W t , ¯ S } and F un til conv ergence is achiev ed. F or a xed RA orien tation matrix F , the subproblem of optimizing the transmit b eamforming matrix W t and the probing-signal cov ari- ance matrix ¯ S can b e reformulated as an SDP and solved using the CVX solvers. F or a xed transmit beamforming matrix W t and probing-signal cov ariance matrix ¯ S , it has b een prov ed in [88] that h T ( F , q U ,k ) w t ,k w H t ,k h ∗ ( F , q U ,k ) in (57b) and h T ( F , q T ,m ) ¯ Sh ∗ ( F , q T ,m ) in (58b) are con vex with resp ect to the amplitude comp onen ts of h ( F , q U ,k ) and h ( F , q T ,m ) , resp ectiv ely . Therefore, the subproblem of optimizing the RA orientation matrix F can b e appro x- imated as a con vex problem and eciently solv ed using the SCA technique. T o illustrate the trade-o b et ween communication and sensing p erformance and to demonstrate the adv antages of RAs in enhancing ISAC capability , we consider an ISA C system with K = 3 users and a circular horizontal sensing region centered at [40 sin π 3 , 40 cos π 3 , − 10] T m with a radius of 5 m. Sp ecically , three users are uniformly distributed in three distinct directions in fron t of the BS at a distance of 50 m, and w e uniformly select M = 8 spatial sampling points within this region. The BS employs a UP A with N = 16 RAs. The maxim um av erage transmit p o w ers allo cated to communication and sensing are set to P max , c = 30 dBm and P max , s = 30 dBm, resp ectiv ely . Fig. 15 shows the minimum received echo p o wer v ersus the required minimum comm unication rate R min for dierent sc hemes. In the baseline system with array-wise rotation, the entire RA arra y shares a common orientation, which is optimized by solving a problem similar to (P-ISA C-D). It is observed that the proposed RA-enabled ISAC system consisten tly achiev es a higher minimum echo pow er than all baseline schemes. This improv emen t stems from the abilit y of individual RAs to indep endently adjust their orien tations/b oresigh ts, enabling the array to recongure … Rotational Directional Diversity Block 1 Block 2 Block M … Data Transmission (a) Dynamic-Orient ation Channel Estimation (b) Fixed-O rientation Channel Estimation Block 1 Block 2 Block M Data Transmission Block m … (2) F (1) F () M F (1) ( 2 ) ( ) ... M    FF F F Fig. 16. T wo representative channel estimation strategies in RA systems: (a) Dynamic-orien tation strategy , and (b) Fixed-orien tation strategy . its directional gain pattern in response to b oth the wireless en vironment and the ISAC p erformance requirements. As a result, the RA-enabled ISA C system signicantly enlarges the ac hiev able trade-o region b et ween comm uni- cation and sensing p erformance. Although the arra y-wise rotation can also enhance the echo p o w er compared with random or xed-orientation designs, its ability to balance comm unication and sensing is fundamen tally limited. This limitation is because individual an tennas cannot b e inde- p enden tly oriented, leading to sub optimal ISAC p erfor- mance compared to the RA-enabled system. Moreo ver, the p erformance gap b etw een the RA-enabled system and the arra y-wise rotation widens as R min increases, highlighting the benets of RA architectures in ISAC applications with stringen t communication and sensing requirements. IV. RA Channel Estimation/Acquisition A ccurate CSI is essential for enabling precise con trol of antenna orientation/boresight and unlo cking the full p oten tial of RA systems. Unlike other exible antenna ar- c hitectures, such as F AS/MA/6DMA that mainly enhance spatial DoF s through antenna repositioning, RA systems ac hieve spatial adaptability b y reorien ting the antenna b oresigh t while keeping the antenna position unchanged. It is worth emphasizing that antenna rotation do es not alter the underlying propagation geometry of the wireless en vironment, such as path loss, angles of arriv al/departure (A oAs/AoDs), or delays. Instead, antenna rotation reori- en ts the radiation pattern, whose shap e is assumed known after antenna c haracterization/calibration. Therefore, for an y given an tenna orientation, the radiation pattern can b e treated as a known directional gain function (e.g., see (11) and (12)), so existing xed-antenna c hannel estimation techniques remain applicable in RA systems after accoun ting for orientation-dependent gains. A. RA Orientation Scheduling for Channel Estimation Dep ending on whether antenna rotation is exploited during the training phase, channel acquisition in RA 20 systems can b e broadly classied into tw o strategies, as illustrated in Fig. 16. • Fixed-orientation channel estimation: RA orienta- tions remain xed within the training interv al. This yields a single-view observ ation and is equiv alent to a con ven tional xed-an tenna system with a given radiation pattern. • Dynamic-orientation channel estimation: RA orien- tations are strategically v aried to collect multi-view measuremen ts, which can improv e estimation accu- racy b y pro viding additional observ ation div ersity . Unless otherwise sp ecied, w e consider a general RA- enabled uplink communication system for ease of ex- p osition, where a BS equipped with an N -element RA arra y serv es K single-an tenna users in the presence of Q scattering clusters. Dene the p osition of user k as q U ,k = [ d k sin ϑ k cos ξ k , d k sin ϑ k sin ξ k , d k cos ϑ k ] T , where d k denotes the distance b et ween the BS’s RA array and user k , and ϑ k and ξ k denote the zenith and azim uth angles, resp ectiv ely . The location of scatterer cluster q , i.e., q C ,q = [ d q sin ˜ ϑ q cos ˜ ξ q , d q sin ˜ ϑ q sin ˜ ξ q , d q cos ˜ ϑ q ] T , can b e similarly dened. Under geometric near-eld propagation, the ov erall m ultipath channel b et w een the BS and user k in (28) can b e reformulated as h k ( F ) = h LoS k ( F ) + h NLoS k ( F ) = ˜ β k, 0 ˜ b k, 0 ( F , q U ,k ) + Q X q =1 ˜ β k,q ˜ b k,q ( F , q C ,q ) , (59) where ˜ β k, 0 = √ β 0 d k and ˜ β k,q = σ q β 0 d q ¯ d k,q denote the propaga- tion co ecients of the LoS and the q -th NLoS components, resp ectiv ely . The eective near-eld array resp onse vector ˜ b k, 0 ( F , q U ,k ) for the LoS c hannel is dened as ˜ b k, 0 ( F , q U ,k ) = diag ( ˜ g U ,k ( F )) 1 2 ˜ a k, 0 ( N , q U ,k ) , (60) where ˜ g U ,k ( F ) = [ g k (  f 1 ) , . . . , g k (  f N )] T denotes the directional gain vector, ˜ a k, 0 ( N , q U ,k ) = [ d k d 1 ,k e − j 2 π λ d 1 ,k , . . . , d k d N,k e − j 2 π λ d N,k ] T denotes the near- eld array resp onse vector with d n,k b eing the distance b et w een RA n and user k . Then, the eective arra y resp onse v ector ˜ b k,q for the NLoS c hannel can b e similarly dened as in (60), i.e., ˜ b k,q ( F , q C ,q ) = diag ( ˜ g C ,q ( F )) 1 2 ˜ a k,q ( N , q C ,q ) . (61) where ˜ g C ,q ( F ) = [ ˜ g q (  f 1 ) , . . . , ˜ g q (  f N )] T denotes the di- rectional gain vector with resp ect to scatterer cluster q , and ˜ a k,q ( N , q C ,q ) denotes the array response vector with respect to scatterer cluster q . Note that when the users and scatterers are lo cated in the far-eld region of the BS array , the distance v ariation across antennas b ecomes negligible in amplitude but appro ximately linear in phase. Thus, the directional gain v ector ˜ g U ,k ( F ) reduces to g k ( F ,  q U ,k ) as shown in (22) or (23). Moreov er, the near-eld array resp onse vector ˜ a k, 0 ( N , q U ,k ) reduces to a far-eld steering v ector a k ( N ,  q U ,k ) in (19) that only dep ends on the direction of user k , i.e.,  q U ,k = [sin ϑ k cos ξ k , sin ϑ k sin ξ k , cos ϑ k ] T . During the uplink training phase, the users transmit m utually orthogonal pilots o ver T a time slots, based on whic h the BS estimates the key channel parameters for eac h user, e.g., propagation co ecien ts A k ≜ { ˜ β k,q } Q q =0 and angular parameters B k ≜ { ( ϑ k , ξ k ) , ( ˜ ϑ q , ˜ ξ q ) } Q q =1 , and then reconstructs the full RA channel { h k } K k =1 . Moreov er, for user k , w e dene the channel parameters inherent in h k as C k = {A k , B k } . In the follo wing, we presen t tw o main estimation schemes for RA systems, with an emphasis on their main dierences regarding the exploitation of the RA orien tation during each channel training p erio d. 1) Dynamic-Orien tation Channel Estimation: In this sc heme, the RA array dynamically adjusts an tenna orien- tations/b oresigh ts across dierent pilot sym b ols to acquire m ulti-view observ ations. Let x ( t ) k denote the pilot symbol transmitted by user k in the t -th slot. Accordingly , the receiv ed signal at the BS is giv en b y y ( t ) = K X k =1 h k  F ( t )  x ( t ) k + n ( t ) , (62) with t = 1 , . . . , T a , where F ( t ) is the RA orien tation matrix in slot t , and n ( t ) ∼ N c ( 0 , σ 2 I N ) denotes A W GN at the BS. In principle, F ( t ) can b e up dated at each slot to increase orientation div ersity . How ever, frequen t switching in tro duces control latency , calibration and feedback ov er- head, and higher complexit y for multi-view data fusion. T o reduce this ov erhead, a blo ck-wise scanning strategy can b e adopted, where RA orientations remain constant within eac h blo c k and change only across blo c ks. Sp ecically , the T a slots are divided into M blo cks, each with T b = T a / M consecutiv e slots. Let F ( m ) denote the RA orientation matrix in blo c k m with m = 1 , . . . , M . Then, the received pilot signal at the BS during blo ck m is given by y ( t ) m = K X k =1 h k  F ( m )  x ( t ) m,k + n ( t ) m , (63) with t = ( m − 1) T b + 1 , . . . , mT b and m = 1 , . . . , M , where x ( t ) m,k denotes the pilot sym b ol transmitted b y user k in slot t of blo ck m , and n ( t ) m denotes the corresp onding A W GN v ector at the BS. By collecting measuremen ts under dier- en t RA orientations, dynamic-orientation channel estima- tion enables joint pro cessing of multi-view observ ations. Since key channel parameters remain inv ariant to antenna orien tation, these observ ations can b e coherently fused to impro ve estimation accuracy by exploiting the additional spatial DoF s oered by antenna rotation. The p erformance gains, how ever, come at the cost of increased control o ver- head and signal processing complexit y . Ecient scanning strategies and low-complexit y estimation algorithms are therefore essential to fully realize the b enets of dynamic- orien tation channel acquisition. 2) Fixed-Orien tation Channel Estimation: In this sc heme, the RA array maintains xed an tenna orienta- tions/b oresigh ts throughout the training p eriod. Sp ecif- ically , the orientation matrix F is preselected and kept unc hanged o ver the T a pilot slots. Under this setting, 21 4 6 8 10 12 14 16 18 20 22 24 Number of antennas, N 10 -5 10 -4 10 -3 10 -2 10 -1 NMSE Isotropic antenna system Fixed orientation design Random orientation design Proposed RA-enabled design Fig. 17. NMSE versus the n umber of an tennas N for the RA-enabled uplink multi-user communication system. the eective channels { h k ( F ) } K k =1 remain constant during training, and th us the RA array op erates equiv alently to a conv entional xed-antenna array with a deterministic radiation pattern. F or notational simplicity , w e drop the explicit dep endence on F and dene h k ≜ h k ( F ) . The receiv ed pilot signal at the BS is given by y ( t ) = K X k =1 h k x ( t ) k + n ( t ) , t = 1 , . . . , T a . (64) Since antenna orientations/boresights remain xed dur- ing channel training, xed-orientation c hannel estimation relies on a single-view observ ation of the propagation en- vironmen t. The main adv an tages of this approach are lo w implemen tation complexity , negligible orientation-con trol o verhead, and full compatibilit y with existing c hannel es- timation proto cols and algorithms for conv en tional xed- an tenna systems. Nev ertheless, the lac k of orien tation div ersity limits sensing cov erage and reduces estimation accuracy , particularly in rich multipath scenarios. F rom the ab ov e, xed-orientation estimation can b e regarded as a direct application of conv entional xed- an tenna channel estimation, requiring no fundamental mo dication to existing techniques. In contrast, dynamic- orien tation channel estimation introduces observ ation di- v ersity by v arying antenna orientations during training. This benet is illustrated in Fig. 17, which sho ws the normalized mean square error (NMSE) versus the n umber of an tennas N for dierent schemes in an RA-enabled uplink multi-user communication system with K = 3 users. It can be observ ed that the NMSE of all sc hemes generally decreases as N increases, since more antennas pro vide higher spatial resolution and more informative observ ations for parameter estimation. Moreo ver, the pro- p osed RA-enabled design consistently outp erforms the isotropic antenna system, the xed-orientation design, and the random-orientation design. This result indicates that prop erly designed RA orientations can pro vide more informativ e multi-view observ ations and thereb y impro ve c hannel estimation accuracy . In the follo wing, w e fo cus on this new dynamic- orien tation framework and discuss how m ulti-view obser- v ations can be fused for c hannel estimation under dieren t RA system setups and signal processing metho ds. B. Channel Estimation for Dieren t RA System Setups Dieren t RA system setups yield distinct observ ation structures, which directly aect the iden tiability and resolv ability of multipath comp onen ts, as well as the training o verhead and signal processing complexit y . In this subsection, w e discuss RA channel acquisition under t wo represen tative setups, i.e., single-RA and multi-RA array setups, with emphasis on their observ ation mechanisms and ho w multi-view measurements can b e fused for chan- nel estimation. 1) Single-RA Setup: In practical deplo yments, the single-RA setup is attractiv e for lo w-cost and resource- constrained platforms, such as APs and Internet-of- Things (IoT) devices, due to its ligh tw eight hardware and compact structure. Unlike traditional single-antenna transceiv ers with xed radiation patterns that only ob- serv e the environmen t from a static viewp oin t, a single-RA transceiv er can acquire multi-view measurements under dieren t orien tations/b oresigh ts n  f ( m ) o M m =1 . By sweeping the antenna b oresigh t ov er time, the transceiver prob es the en vironment from multiple angular p ersp ectiv es, thereby in tro ducing observ ation diversit y ev en with a single an- tenna. F rom an implementation p ersp ectiv e, channel estima- tion under the single-RA setup can b e realized via ei- ther “passiv e listening” or “active sensing”, dep ending on whether the RA no de passively receiv es known pi- lot signals or actively transmits probing wa v eforms and pro cesses echoes. Let x ( t ) m denote the known transmitted signal in the time slot t within the m -th orientation blo c k: a pilot signal from an external transmitter (passiv e listening) or the probing wa veform from the single-RA no de (active sensing). In passiv e listening, the single- RA no de receiv es pilot symbols broadcast by an external transmitter. Within the m -th orientation blo c k, the RA main tains a xed orientation  f ( m ) for T b slots, and the receiv ed signal in (63) reduces to a scalar form y ( t ) m = h   f ( m )  x ( t ) m + n ( t ) m , (65) with t = ( m − 1) T b + 1 , . . . , mT b and m = 1 , . . . , M . By adjusting the antenna orientations/boresights ov er m ultiple time blo c ks, the receiver collects a sequence of observ ations, from which the channel parameters can b e inferred by jointly pro cessing the accum ulated channel measuremen ts. In activ e sensing, the single-RA no de transmits probing wa veforms and pro cesses echoes. Under a generic discrete-time baseband mo del with ˜ Q resolv able reection paths, the echo signal can be expressed as y ( t ) m = ˜ Q X ˜ q =1 β ˜ q g tx , ˜ q   f ( m )  g rx , ˜ q   f ( m )  e − j 4 π λ d ˜ q x ( t ) m + n ( t ) m , (66) 22 with t = ( m − 1) T b + 1 , . . . , mT b and m = 1 , . . . , M , where β ˜ q and d ˜ q denote the round-trip propagation co ecien t and propagation distance of the ˜ q -th path, resp ectiv ely , and g tx , ˜ q ( · ) and g rx , ˜ q ( · ) represent the trans- mit/receiv e radiation co ecien ts. F or a monostatic single- RA transceiv er, one may set g tx , ˜ q   f ( m )  = g rx , ˜ q   f ( m )  = r G  ϵ (  f ( m ) ,  q ˜ q ) , φ (  f ( m ) ,  q ˜ q )  , with  q ˜ q denoting the direc- tion v ector of the ˜ q -th path. Despite their dierent op erating mo des, b oth paradigms rely on accumulating m ultiple observ ations across time and antenna orientations. By stacking the received signal o ver T a time slots, the single-RA observ ation vector can b e expressed as y ≜ h y (1) 1 , . . . , y ( T b ) 1 , . . . , y ( M − 1) T b +1 M , . . . , y ( M T b ) M i T ∈ C T a × 1 , (67) with T a = M T b . Even with a single RF chain, the single-RA system enables multi-persp ectiv e spatial ob- serv ation via an tenna rotation, reducing hardware cost while improving multipath resolv ability . Such a setup is simple and cost-eectiv e, yet it often suers from limited view diversit y and constrained parameter resolv ability under practical training constraints. Sp ecically , since the spatial information is acquired sequentially via b oresight sw eeping, the n umber of orien tations within one coherence blo c k is b ounded by switching latency and channel v ari- ation. In rich-scattering environmen ts, w eak paths ma y b e obscured by dominant components unless sucien tly dense sampling is emplo yed, which inevitably increases the pilot o verhead and training latency . 2) Multi-RA Array Setup: In contrast to the single-RA setup that observes spatial information sequentially via b oresigh t sweeping, the multi-RA array acquires spatial information ov er multiple an tennas in parallel, pro ducing an N -dimensional snapshot p er training slot. By concate- nating snapshots under a con trolled orien tation sc hedule, one forms a spatio-temp oral observ ation matrix that en- ables more reliable multipath resolv ability and parameter iden tiability than sequen tial single-RA acquisition. As N increases, the eective array ap erture and spatial DoF s gro w, thereby enhancing angular discrimination and the abilit y to resolv e closely spaced multipath comp onents. Consider a multi-RA arra y with N directional antennas arranged in a general geometry . At the t -th training slot, the receiv ed signal follows the RA c hannel model in (28), yielding an N -dimensional observ ation v ector y ( t ) ∈ C N × 1 . Collecting received signals ov er T a training slots giv es the spatio-temp oral observ ation matrix: Y ≜ h y (1) 1 , . . . , y ( T b ) 1 , . . . , y ( M − 1) T b +1 M , . . . , y M T b M i ∈ C N × T a , (68) whic h jointly captures spatial array responses across an tennas and temp oral ev olution induced by antenna rotation. Ob viously , as N increases, the multi-RA setup gener- ates high-dimensional m ulti-view observ ations, providing substan tially more information for multipath resolution and parameter estimation. How ever, fully exploiting these gains requires carefully designed lo w-o verhead estimation sc hemes. On the one hand, the dimensionalit y of orien- tation congurations grows rapidly with N , making join t orien tation design c hallenging under practical con trol and latency constraints. On the other hand, high-dimensional observ ations demand adv anced spatio-temporal signal pro- cessing to leverage geometric priors and correlations for c hannel extrap olation and parameter inference. Promising approac hes include structured orientation control (e.g., hierarc hical co deb ook scanning, adaptive b oresight re- nemen t) com bined with spatio-temp oral estimation al- gorithms such as subspace/parametric tting [84], [92], sparse reconstruction [93]–[95], and tensor-based factor- ization [96] to exploit lo w-rank or sparse structures in Y . Moreov er, AI-based methods, such as neural netw orks [97], can b e emplo yed to capture correlations across orien tations and appro ximate the nonlinear mapping from training data to key c hannel parameters, enabling ecient c hannel estimation in complex environmen ts. C. Signal Pro cessing Metho ds for RA Channel Estimation In this subsection, we present a structured ov erview of ecien t RA channel estimation schemes based on v arious signal pro cessing metho ds, including maximum-lik eliho od (ML), subspace-based, compressed sensing, beam training, and machine learning approaches. W e elab orate on their fundamen tal principles for CSI acquisition and discuss how RA orientation control can b e exploited to improv e esti- mation accuracy under practical training and complexity constrain ts. In the follo wing, we consider a typical user k and aim to estimate h k . F or notational simplicity , the subscript k is omitted whenev er possible. 1) ML-Based Estimation: As demonstrated by the geometric c hannel mo del in (59), the RA channel can b e parameterized by a set C k that collects dominant path parameters, e.g., AoA pairs and propagation co ecients. A key prop ert y of RA systems is that antenna orien ta- tion recongures the radiation pattern or array manifold but do es not alter the underlying propagation geometry within a coherence interv al. Hence, a common practice is to estimate the sparse channel parameters and then reconstruct the CSI for arbitrary orientations based on these parameters. Consider the received signal mo del in (63) for the dynamic-orien tation case. By stacking the received signal v ectors { y ( t ) } T a t =1 , the received signal in (68) can b e rewritten as follows: y ≜ v ec ( Y ) = ˜ S ( F ; η ) β + n , (69) where ˜ S ( F ; η ) ∈ C N T a × ( Q +1) denotes the observ ation ma- trix depending on RA conguration F and pilot sequence { x ( t ) } T a t =1 , β = [ ˜ β k, 0 , ˜ β k, 1 , ..., ˜ β k,Q ] T denotes the propaga- tion co ecien t vector, η = [ ϑ k , ˜ ϑ 1 , ..., ˜ ϑ Q , ξ k , ˜ ξ 1 , ..., ˜ ξ Q ] T collects angular parameters, and n ∈ C N T a × 1 stac ks the A W GN vectors. The ML estimator seeks β and η that maximize the likelihoo d function P ( y | β , η ) . 23 Based on the received signal in (69), the log-lik eliho o d function is given by ln ( P ( y | β , η )) = − N T a ln ( σ 2 π ) − 1 σ 2    y − ˜ S ( F ; η ) β    2 2 . (70) Th us, ML-based estimation is realized by solving the follo wing optimization problem: { η ⋆ , β ⋆ } = arg min η , β   y − ˜ S ( F ; η ) β   2 2 . (71) When multi-view training data is sucien tly informa- tiv e, ML estimation is statistically ecient and can approac h the CRB. Ho w ever, (71) is generally a high- dimensional non-conv ex problem, making joint optimiza- tion ov er { η , β } computationally prohibitiv e. This moti- v ates ecient techniques suc h as alternating minimization (e.g., BCD ov er η and β ). The p erformance of ML-based estimation is also in- uenced b y the RA orientation schedule F in ˜ S ( F ; η ) . Rep eated or highly similar orien tations yield correlated measuremen ts and an ill-conditioned sensing matrix, am- plifying noise and causing parameter coupling, thereb y requiring more pilots to achiev e a desired target estima- tion accuracy . Therefore, a fav orable orientation sched- ule should ensure div erse angular cov erage and av oid redundan t views within the feasible orientation region. F or example, when RAs are constrained to steer within a spherical cap as in (38), naive uniform sampling in the angular domain leads to non-uniform cov erage: grid p oin ts cluster near the p ole and b ecome sparse near the b oundary . T o achiev e near-uniform sampling, spherical Fib onacci sampling can b e employ ed to generate reference directions with improv ed angular uniformity . 2) Subspace-Based Metho d: In nite-scattering scenar- ios, the n umber of eective channel paths ( Q + 1) is t ypi- cally muc h smaller than the num b er of RAs N . This nite set of dominant propagation paths induces a low-rank structure in the RA channel, which can be exploited b y subspace-based estimation methods. Specically , classical algorithms such as MUltiple SIgnal Classication (MU- SIC) and Estimation of Signal P arameters via Rotational In v ariance T echniques (ESPRIT) [92] can b e applied to estimate the angular parameters η , follow ed by a least- squares (LS) estimation of the propagation co ecients β [84]. Based on the geometric channel mo del in (59), the RA c hannel v ector can be rewritten as follo ws: h ( F ) = A ( F ; η ) β , (72) where A ( F ; η ) = [ ˜ b k, 0 , ˜ b k, 1 , . . . , ˜ b k,Q ] ∈ C N × ( Q +1) de- notes the RA arra y manifold. By exploiting T b pilot slots within block m under a giv en RA orien tation F ( m ) , the sample co v ariance matrix of the received pilot signals is giv en by ˆ R m = 1 T b T b X t =1 y ( t ) m  y ( t ) m  H . (73) Eigen v alue decomp osition of ˆ R m yields the estimated signal-subspace and noise-subspace matrices, denoted by E s ,m ∈ C N × ( Q +1) and E n ,m ∈ C N × ( N − Q − 1) , respectively . The MUSIC algorithm can then b e applied to construct the pow er spectrum V m ( ϑ, ξ ) = 1 ˜ b H ( F ( m ) ; ϑ, ξ ) E n ,m E H n ,m ˜ b ( F ( m ) ; ϑ, ξ ) , (74) whose p eaks reveal the angular parameters of the dom- inan t paths in blo c k m . Since the ph ysical AoAs/A oDs are common across blo cks, information from dierent RA orientations can b e combined by fusing their p o wer sp ectra. A simple yet eectiv e approach is to a verage the blo c k-wise spectra as follo ws: ¯ V ( ϑ, ξ ) = 1 M M X m =1 V m ( ϑ, ξ ) . (75) The ( Q + 1) largest p eaks of ¯ V ( ϑ, ξ ) are then selected as the nal angular estimates ˆ η . Finally , b y stac king all T a = M T b pilot observ ations into a single vector y as in (69), the LS estimate of the propagation co ecients is obtained as ˆ β =  ˜ S H ˜ S  − 1 ˜ S H y . (76) Substituting ˆ η and ˆ β in to (72) yields a parametric reconstruction of the RA c hannel. 3) Compressed Sensing (CS)-Based Estimation: Due to the inheren t sparsity of the nite channel in (59) within the angular (and distance) domain, CS-based metho ds can b e eectiv ely applied for RA channel acquisition. These methods aim to recov er key channel parameters from a reduced num b er of pilot observ ations, thereby lo wering training o verhead and computational complexity . The core idea of CS-based estimation is to discretize the con tinuous parameter space into a nite grid and represen t the channel as a sparse combination of dictionary atoms. Let α ∈ C J × 1 denote the sparse co ecien t vector dened on a grid of size J . By stac king pilot observ ations collected under multiple RA orientations, the received signal mo del can be formulated as a linear sparse mo del: y = Φ ( F ) α + n , (77) where Φ ( F ) ∈ C N T a × J is the sensing matrix constructed from the eective RA arra y manifold asso ciated with orien tation F and the known pilots. In this con text, CS- based RA channel estimation reduces to a sparse reco very problem: min α k α k 0 s.t. k y − Φ ( F ) α k 2 2 ≤ ϵ, (78) where ϵ denotes the error tolerance. While k · k 0 mini- mization is generally NP-hard, it can b e ecien tly ap- pro ximated using practical sparse recov ery algorithms. In particular, classical CS algorithms, suc h as orthogonal matc hing pursuit (OMP) [93] and compressiv e sampling matc hing pursuit (CoSaMP) [94], [95], can b e emplo yed to estimate the channel parameters corresponding to the 24 columns of Φ ( F ) with non-zero coecients in α . Dieren t from conv entional xed-antenna arra ys, a distinctiv e feature of RA systems is that the sensing matrix Φ ( F ) is orientation-dependent. The RA orien tation conguration reshap es the eectiv e radiation patterns and directly impacts the mutual coherence and conditioning of Φ , whic h are critical to sparse recov ery p erformance. Sp ecically , v arying orien tations across RAs inuence the correlation among dictionary atoms, and p oorly designed orien tation schedules ma y degrade recov ery accuracy . Therefore, it is essential to design orientation-a ware sens- ing matrices whose columns join tly enco de b eamforming directions and antenna orientations. 4) Beam T raining-Based Estimation: Beam training is a practical alternativ e for acquiring implicit CSI in sparse channels by searching ov er a nite set of candidate b eam/orien tation congurations. Dieren t from mo del- based estimation that explicitly recov ers the geometric parameters η and β , beam training aims to select the b est transmission mode (determined co dewords) that maximizes the received signal p o wer or SNR, which is particularly app ealing when low-complexit y link align- men t is the primary ob jective. In RA systems, mo de selection is jointly gov erned by the RA orientation con- guration, whic h reshapes the radiation pattern, and the con ven tional transmit/receiv e b eamforming weigh ts. Let F = { F (1) , . . . , F ( |F | ) } denote the RA orien tation co de- b ook and W = { w (1) , . . . , w ( |W | ) } denote a conv entional b eamforming/com bining co debo ok. Then, a typical beam- training problem in an RA system can b e formulated as max F , w | w T h ( F ) | 2 s.t. F ∈ F , w ∈ W . (79) A ccordingly , b eam training is equiv alent to selecting the co dew ord pair ( F ⋆ , w ⋆ ) that yields the strongest eective c hannel gain. In addition, the co deb ooks F and W need to be designed to satisfy compactness and scalability . Sp ecically , the co debo ok size should be reduced as muc h as p ossible to reduce the beam training ov erhead. Dieren t from xed antenna congurations, RA in tro duces an additional controllable dimension, i.e., the b oresight rota- tion, which changes the eective manifold and may alter the distinguishability b etw een dierent b eams. There- fore, practical RA beam training critically depends on the orientation-a ware co debo ok design and lo w-ov erhead mo de selection strategies. The RA orientation co deb ook F should provide sucient cov erage of the b oresigh t domain and ensure go od separability among co dew ords. T o satisfy the cov erage requiremen t, a baseline idea is to discretize the b oresigh t domain into ( ϑ, ξ ) grids and map each grid point to an RA orien tation F ( i ) . F or near- eld/wideband settings, F ( i ) can further incorp orate a fo cal-distance bin, yielding a multi-resolution co deb ook. In addition, RA orientation induces v arying radiation patterns across antennas that disrupt the orthogonality of con ven tional co deb ooks. Therefore, it is essen tial to dev elop radiation-pattern-a ware co debo oks where each co dew ord jointly enco des a b eamforming direction along with sp ecic antenna orien tation congurations, thus impro ving codeword distinguishabilit y . In this case, W can follo w standard designs, e.g., discrete F ourier transform (DFT) co deb ooks [98], while F is tailored to RA features and radiation-pattern characteristics. Giv en F and W , b eam training p erforms mo de se- lection by sw eeping a subset of candidate codeword pairs  F ( i ) , w ( j )  , measuring the received p ow er metric O ( i, j ) ≈ | ( w ( j ) ) T h ( F ( i ) ) | 2 , and selecting the maximal v alue in ( 79). A straigh tforward approac h is to conduct an exhaustive search ov er all p ossible co deb ooks. How- ev er, this may incur prohibitively high training ov erhead, esp ecially when the co deb ook size |F ||W | is large. T o reduce the training o verhead, hierarchical (coarse-to-ne) scanning is commonly adopted: a coarse stage uses wide b eams (and coarse orientation sectors) to lo calize a small candidate set, follow ed by rened probing with narrow er b eams within the selected sector to iden tify the optimal index ( i ⋆ , j ⋆ ) . 5) Learning-Based Channel A cquisition: Compared with traditional estimation schemes that rely on accurate mathematical mo dels, learning-based metho ds provide a data-driv en alternativ e that can incorp orate multi-view information to impro ve estimation accuracy . The key idea is to learn a nonlinear mapping from the observ ed pilot data (input) to a compact representation of the c hannel (output), and then reconstruct the CSI for subsequent data transmission. This paradigm is particularly attractiv e for RA systems b ecause antenna orientations reshap e the eectiv e radiation pattern and thereby generate infor- mativ e m ulti-p ersp ectiv e measurements. Suc h multi-view div ersity can b e leveraged to reduce pilot ov erhead and enhance robustness against model mismatches, calibration errors, and hardware impairments. T o facilitate standard learning arc hitectures, we stac k all m ulti-view observ ations in (68) as y = vec( Y ) ∈ C N T a × 1 and con vert them in to a real-v alued input: ˜ y = [ <{ y } T , ={ y } T ] T ∈ R 2 N T a × 1 . (80) Note that in nite-scattering environmen ts, the RA c han- nel can b e well characterized using some key parameters (e.g., χ = [ β T , η T ] T ∈ R 3( Q +1) × 1 ). Hence, instead of di- rectly estimating the full CSI, a more ecien t and common practice is to estimate the lo w-dimensional parameters and then reconstruct the channel. Accordingly , learning-based RA channel estimation constructs a nonlinear mapping b et w een the input data (multi-view observ ations) and the output data (key channel parameters), which is given by ˆ χ = f ξ  ˜ y ; { F ( m ) } M m =1  , (81) where f ξ ( · ) : R 2 N T a × 1 7→ R 3( Q +1) × 1 denotes a training mo del parameterized by ξ . Moreov er, the training data can b e enriched b y using dierent orientation congurations, thereb y improving view diversit y and facilitating feature learning. Learning-based metho ds also oer several adv antages 25 o ver model-based techniques in RA systems. Sp eci- cally , the data-driven/model-free structure of a learning- based approach can provide robustness against measure- men t/mo del impairmen ts, facilitate feature extraction and adaptation to en vironmental changes, and enable low- complexit y inference compared with iterative optimiza- tion. More imp ortan tly , RA rotations naturally provide m ulti-view pilot measurements, enabling the learning- based estimator to fuse complemen tary p ersp ectiv es and reco ver a compact set of geometric c hannel parame- ters with reduced pilot ov erhead and improv ed track- ing capability . A v ariet y of learning mo dels, ranging from the light weigh t multila yer perceptrons (MLPs) to deep architectures suc h as conv olutional neural netw orks (CNNs) and transformers, can b e employ ed [99]. De- p ending on label a v ailability and deploymen t constrain ts, one may adopt sup ervised learning (SL) using lab els generated by mo del-based estimators or sim ulations, self- sup ervised/unsupervised learning via physics-consisten t reconstruction losses, or federated learning (FL) to train a shared estimator across distributed devices with limited data sharing (e.g., [100], [101]). V. RA Conguration and Deplo yment F rom concept to implementation, RA conguration and deplo yment play a critical role in determining b oth system p erformance and hardware eciency . While RA oers a compact and scalable solution to enhance spatial exibilit y , its practical conguration and deplo yment re- main challenging due to stringent hardw are and control constrain ts. Key design asp ects—suc h as b oresigh t control metho ds, rotational range and granularit y , arra y struc- ture, and deploymen t strategy—directly aect control accuracy , latency , size, and cost, thereb y impacting ov erall p erformance and in tegration complexity . In the following, w e discuss representativ e RA congurations/deploymen ts, highligh t their resp ectiv e adv antages and disadv antages, and provide guidelines for selecting suitable RA congu- rations/deplo yments in practice. A. Mechanical vs. Electronic Rotation F rom a hardw are p erspective, the RA boresight can b e recongured through either mechanically-driv en or electronically-driv en mechanisms, as depicted in Fig. 1 . Both approaches enable eective b oresigh t control but exhibit dierent trade-os in accuracy , agility , complexity , and cost. F or mechanically-driv en rotation, each antenna/arra y is moun ted on an actuation platform (e.g., a gim bal or serv o- driv en stage) to physically steer the b oresigh t. A typical realization inv olves mounting the RA on a servo-motor- con trolled platform, where platform rotation directly ad- justs antenna orientation in 3D space [102], [103]. Alter- nativ ely , miniaturized electro-mec hanical actuation based on MEMS can facilitate compact in tegration and simplify deplo yment [27]. Mec hanical rotation generally provides a wide steering range and ne angular resolution, but at the cost of non-negligible actuation latency and p oten tial mec hanical wear. F or instance, MEMS-based actuators t ypically consume milliwatt-lev el pow er and exhibit re- sp onse times from µs to ms. In practice, mechanically- driv en metho ds are constrained b y actuator complexity , main tenance requiremen ts, and ph ysical limitations suc h as friction and limited lifetime under rep eated rotation. F or electronically-driven rotation, the an tenna remains ph ysically xed, while its b oresight is recongured by electronically reshaping the excitation or imp edance dis- tribution, thereby emulating radiation pattern rotation without moving parts. Representativ e designs include (i) recongurable feeding netw orks or multi-feed radiators, where switching among feed p oin ts steers the mainlob e [104]; and (ii) parasitic-element-based designs with elec- tronically tunable loads (e.g., v aractor or PIN diodes), where induced curren ts and mutual coupling are adjusted to rotate the pattern [105]. Since practical feed/load congurations are nite, the ac hiev able steering directions are typically discrete and predened. T o enable quasi- con tinuous adjustmen t, tunable materials and recong- urable metasurfaces can b e employ ed [106], though they require sophisticated biasing and calibration and may face bandwidth or thermal stability challenges. Ov erall, electronically-driv en rotation is more compatible with compact platforms and can ac hieve reconguration la- tencies ranging from ns to ms, dep ending on sp ecic implemen tation. Nevertheless, b ecause b oresigh t rotation is realized via excitation/imp edance reconguration, the syn thesized patterns across electronic states are generally not exactly rotated replicas and may exhibit v ariations in mainlob e width, sidelob e structure, p olarization, and phase cen ter. These tw o approaches are complementary: mechani- cal rotation provides wide-angle coarse steering, whereas electronic rotation enables low-latency ne adjustmen t. Hybrid RA arc hitectures can integrate b oth, with me- c hanical rotation handling infrequen t large-angle up dates and electronic control compensating for residual p oin ting errors and supp orting fast b eam trac king. B. Contin uous vs. Discrete Rotation Ideally , eac h an tenna orien tation/b oresigh t can b e adjusted contin uously , allowing arbitrarily ne angular resolution. The contin uous feasible set of orientations is mo deled as in (37). In compact arra y deploymen ts, ho wev er, large rotations may increase mutual coupling b et w een adjacen t antennas and induce pattern distortion. T o accoun t for such eects and restrict excessive deviation, an additional rotational constraint can b e imp osed as in (38). This con tinuous orien tation mo del pro vides theo- retical p erformance limits of RA-enabled systems under practical constrain ts [23], [24]. While contin uous adjustmen t is b enecial for optimizing comm unication and sensing p erformance, it is dicult to implemen t in practice b ecause high-resolution rotation requires higher cost and more complex hardw are design. 26 As a cost-eective alternativ e, antenna orientation can b e implemen ted discretely , where each RA steers its boresight to ward a nite set of candidate directions. Let I ψ , I θ , and I ϕ denote the num b ers of quantization levels for rotation angles ψ , θ , and ϕ , resp ectiv ely . Then, the corresponding discrete angle sets are F ′ X =  ¯ X 1 , ¯ X 2 , . . . , ¯ X I X  , (82) with X ∈ { ψ , θ, ϕ } , where 0 ≤ ¯ X 1 < · · · < ¯ X I X < 2 π . The resulting discrete orientation co debo ok is the Cartesian pro duct F ′ = F ′ ψ × F ′ θ × F ′ ϕ , whose s ize scales as |F ′ | = I ψ I θ I ϕ . As a sp ecial case, uniform quantization discretizes each angle into evenly spaced grids. F or ex- ample, F ′ ψ = { ψ i } I ψ i =1 with ψ i = ψ 1 + ( i − 1)∆ ψ , where ∆ ψ is the quan tization step size; F ′ θ and F ′ ϕ can b e dened similarly . Finite-resolution control introduces a fundamen tal p erformance-complexity trade-o: increasing { I ψ , I θ , I ϕ } improv es steering accuracy but enlarges the co debo ok size and feedbac k ov erhead, whereas fewer lev els reduce cost at the expense of p erformance. Moreov er, discrete rotation complicates orien tation optimization, since it in tro duces discrete v ariables that are generally harder to handle than con tinuous ones [86]. C. Sparse vs. Non-Sparse Arra y Considering practical constraints on cost, hardw are complexit y , and deplo yment space, the RA arra y structure needs to be carefully designed to balance sensing and com- m unication p erformance across dierent wireless systems. The arra y geometry determines the physical ap erture and spatial sampling pattern, which in turn shap es the mainlob e width, sidelob e level, and grating-lob e b ehavior, ultimately impacting ac hiev able p erformance. Sp ecically , there are t wo representativ e RA arra y structures: non- sparse (compact) and sparse RA arrays. A compact arra y t ypically emplo ys approximately half-wa velength in ter-antenna spacing, pro viding dense spatial sampling that enables stable b eam steering ov er a broad angular range while suppressing grating lob es. Such lay outs also facilitate in tegrated packaging and calibration, making them attractive for platforms with limited size. Ho wev er, densely pac king antennas increases hardw are cost, energy consumption, and signal pro cessing o verhead. Moreov er, for a xed num b er of antennas, compact arrays hav e a relativ ely limited physical ape rture, restricting spatial res- olution and interference suppression. In addition, compact deplo yments are also more susceptible to mutual coupling and near-eld in teractions, and RA b oresigh t rotation ma y cause pattern o verlap among adjacent an tennas, further distorting the eective array resp onse. T o o vercome these limitations, sparse RA arra ys ha ve emerged as a promising architecture that enlarges the ap erture without increasing the num b er of antennas. By relaxing the half-wa velength spacing constraint, sparse arra ys allow greater inter-an tenna separation. Such sparse geometries yield sharp er b eams and ner spatial resolu- tion, desirable for high-resolution sensing and improv ed user separability/in terference suppression in ISAC sys- tems [107]. Meanwhile, increased antenna spacing also alleviates m utual coupling and reduces pattern ov erlap during antenna rotation, thereb y enhancing recongura- tion exibility . F urthermore, certain sparse geometries enable dierence/sum co-arrays, creating enlarged virtual ap ertures and additional sensing DoF s, which are particu- larly benecial for lo calization and m ulti-target resolution. Nev ertheless, sparse sampling raises the risk of elev ated sidelob es and grating lob es, whic h ma y cause energy leakage and severe in ter-user interference when users fall in to o verlapping grating-lob e regions. In sensing, they ma y also induce angular am biguities. T o address this issue, sparse RA arrays can jointly exploit ap erture gain from an tenna placement and radiation pattern recongurabilit y from orien tation/b oresigh t control. Specically , sparse placemen t sharp ens the mainlob e via an enlarged ap erture, while RA rotations reshap e radiation patterns to regu- late sidelob es/grating lobes and provide view div ersity . This joint design is particularly app ealing for ISAC, as orien tation diversit y enhances iden tiability and clutter robustness in sensing while enabling interference shaping and impro ving link reliability in communication, without sacricing the ap erture adv antage of sparse arra ys. D. Distributed vs. Centralized Deploymen t While RA hardware and orien tation control deter- mine the lo cal radiation b eha vior of each antenna/arra y , deplo yment across the netw ork dictates global spatial div ersity and co ordination ov erhead. The choice of de- plo yment strategy for RAs should b e tailored to system requiremen ts, striking a balance b et ween hardware cost and p erformance. At the netw ork level, there are tw o main strategies in RA deplo yment: centralized deplo y- men t, where an tennas/arrays are co-lo cated on a single platform to facilitate coordinated control and coherent pro cessing [83], [85], [88], [90]; and distributed deploymen t, where an tennas/arrays are placed across m ultiple spatially separated nodes to enhance cov erage and pro vide macro- div ersity gains against shadowing and blo ckage [81], [82]. Deplo yment strategies of RAs signicantly impact ef- fectiv e channel realizations and thus the fundamental p erformance limits of RA-enabled systems. Centralized deplo yment is generally fav orable for coheren t beam- forming with tigh t sync hronization, whereas distributed deplo yment oers lo cation diversit y but requires more stringen t sync hronization, CSI exchange, and calibration. Implemen tation considerations include operational cost, user/target distribution, space constraints, and propa- gation environmen t. F or instance, in rich-scattering or sparsely p opulated environmen ts, distributed deplo yment impro ves co verage via macro-div ersity and mitigates shad- o wing/blo c kage. By contrast, aerial or spaceb orne plat- forms, often constrained by size, weigh t, and p o wer (SW aP), fav or centralized compact arrays with small fo otprin ts and electronically-driv en or MEMS-actuated rotation for practical integration. Moreov er, these plat- forms are also sub ject to mechanical vibrations, pointing 27 Rotatable Antenna Servo Zenith Angle Azimuth Angle Directional Antenna Radiation Pattern TX RX Laser Radar TX R X Radiation Pattern Camera (a) (b) Radar Sensing/ V isual Recognition RX Position RX USRP USRP TX Controller PC RA Antenna Deployment Signal T ransmission PC PC Isotropic Antenna Servo Control Serial Port Communication Antenna Deployment Position Information (c) Laser Radar Camera (d) Fixed-antenna RA Fig. 18. Prototypes of sensing-assisted RA. (a) Visual recognition [102]. (b) Radar sensing [103]. (c) Prototype architecture. (d) The received SNR versus azimuth angle of the RX and constellation diagrams. jitters, and thermal v ariations, which can induce p oin ting errors and distort the eective array resp onse, thereb y degrading performance [108]. T o alleviate these eects, RA deploymen ts should incorp orate ruggedized hardware, robust pac kaging, and calibration-aw are con trol. VI. RA Prototypes and Related Pro ducts This section provides an o verview of existing protot yp es that employ dierent implementation strategies for RA- enabled wireless sensing and communication, as well as their exp erimen tal results v alidating the performance gains practically achiev able through antenna rotation. F urthermore, we presen t several representativ e commer- cial pro ducts whose design concepts align with the core principles of RA, thereby demonstrating the feasibilit y and p oten tial of RA-like solutions in practical deplo yments. A. Single RA Prototypes T o v alidate the feasibility of single-RA implementations, Fig. 18 presen ts tw o represen tative sensing-assisted RA protot yp es and corresp onding exp erimen tal results. In these prototypes, the core concept is to detect the spatial p osition of the receiv er and conv ert it into the correspond- ing azimuth and elev ation angles of the RA, enabling the serv o system to precisely rotate the RA boresight direction to ward the target. T o acquire target/receiver p osition in- formation, t wo representativ e sensing mo dalities are visual recognition and radar sensing. In the visual recognition- assisted RA prototype sho wn in Fig. 18(a) [102], a camera and a p ersonal computer (PC) jointly form a vision mo dule that detects and tracks the receiver as a visual target. Moreov er, the “Y ou Only Lo ok Once” (YOLO) net work [109] is emplo yed for real-time target detection, while the “DeepSOR T” algorithm [110] is adopted to trac k the mov ement of the receiver o ver time. Therefore, the spatial p osition of the receiver is captured within the image and serves as guidance for subsequent an- tenna b oresigh t adjustmen t. Mean while, the radar sensing- assisted RA prototype illustrated in Fig. 18(b) acquires the receiver’s spatial p osition information via radar de- tection. Sp ecically , a laser radar mo dule mounted at the transmitter scans the surrounding environmen t and estimates the lo cation of the receiv er, using a time-of-igh t (T oF) approac h. Despite the dierent sensing modalities, b oth implementations rely on a common con trol pro cess. Through geometric transformation, the acquired p osition information is con verted in to target azimuth and elev ation angles, which are processed by a micro con troller (e.g., STM32) to driv e a servo motor via a proportional-integral- deriv ative (PID) algorithm. This closed-lo op mec hanism dynamically steers the antenna to comp ensate for receiver mobilit y , thereb y sustaining precise b oresigh t alignmen t. In the exp erimen t, the sensing-assisted RA implemen- tations operate at a carrier frequency of 5.8 GHz and emplo y 16-quadrature amplitude mo dulation (16-QAM), with a transmit p ow er of 10 dBm and a data rate of 2 Mbps. As shown in Fig. 18(d), b oth sensing-assisted RA implementations achiev e a notable p erformance gain compared with the con ven tional xed-antenna system. In particular, an SNR improv ement of appro ximately 7 dB is observ ed, accompanied by a signican tly clearer 16- QAM constellation diagram. These exp erimental results conrm that accurate b oresigh t adjustment enabled b y en vironmental sensing eectiv ely enhances link quality , v alidating the practicality of sensing-assisted RA systems. B. RA Array Prototype While the aforemen tioned prototypes v alidate the ef- cacy of the single RA, practical BSs t ypically employ an tenna arrays to achiev e high directional gain and spatial m ultiplexing. T o v alidate the practical feasibility of the RA arra y , Fig. 19 illustrates its deploymen t in tw o represen- tativ e scenarios. As shown in Fig. 19(a), the RA array en- ables high-gain p oin t-to-p oin t directional communication in indo or scenarios, such as AR/VR services, b y dynami- cally adjusting the b oresight direction of each an tenna [26]. Sp ecically , by accurately perceiving and trac king the target user’s lo cation, the system can orient the RAs to ward the desired direction, thereby concentrating signal 28 TX Radiation Pattern RX RA Array RA element An emulated low-altitude ISA C scenario between a UA V and a BS equipped with an RA array RA array (at emulated BS) An emulated dir ectional point-to-point communication scenario Component Directional antenna Servo-motor gimbal Acrylic plate FPGA PWM controller Power adapter T otal Number × 4 × 4 × 1 × 1 × 1 × 1 -- Cost (USD) $81.7 $19.08 $4.18 $58.35 $3.76 $4.18 $171.25 Rotation FPGA RA Array (a) (b) Fig. 19. Prototype of RA array in dierent scenarios. p o w er and supp orting high-rate, lo w-latency transmission. In such scenarios, the RA array eectively exploits spatial DoF s to improv e link reliability and throughput, thereby realizing perception-assisted directional comm unication. F urthermore, the RA array can also b e eectiv ely applied in low-altitude ISAC sce- narios as illustrated in Fig. 19(b). A short demonstration video is a v ailable on Y ouT ub e at h ttps://www.youtube.com/watc h?v=L1aMV aGj5rw. In this exp erimen t, the terminal emulator is deplo yed on an unmanned aerial vehicle (UA V) platform to em ulate the lo w-altitude scenario. Moreo ver, the RA arra y op erates as a BS, supp orting real-time sensing and communication with the UA V. In particular, the RA array can accurately e stimate the AoA of incoming signals and track the UA V’s tra jectory in real time. F urthermore, the system maintains b oresigh t alignment ev en during UA V maneuv ers such as climbs, turns, and ho vers. Consequently , the comm unication link remains stable, and the sensing results accurately track the UA V’s motion. This dynamic trac king capabilit y v alidates the robustness of the RA array in managing low-altitude mobile targets and underscores its strong p oten tial as a foundational infrastructure for future 3D netw ork co verage. Notably , the RA array prototype demonstrates remark- able cost-eciency . As summarized in Fig. 19, the total hardw are cost is only 171.25 USD, including the direc- tional antenna, servo gimbal, eld-programmable gate arra y (FPGA), and control mo dules. This highly com- p etitiv e cost structure not only highlights the practicality and eciency of the RA architecture but also provides a cost-eective and scalable solution for the large-scale deplo yment of dense lo w-altitude ISA C netw orks. C. Related Commercial Pro ducts Bey ond these academic prototypes, the concept of dynamic antenna orien tation/b oresigh t has also b een adopted in recent commercial industrial pro ducts, fur- ther v alidating its practical v alue. F or instance, TP- Link has launched the Archer AXE200 Omni, whic h is a Wi-Fi 6 extended (Wi-Fi 6E) router equipp ed with smart mechanically-driv en antennas, as shown in Fig. 20(a) [111], [112]. Specically , this device emplo ys motorized mec hanisms to dynamically reorient its an ten- nas according to the spatial distribution and mov ement of connected devices. By tracking user lo cations and adapting to the lay out of the environmen t, the router mec hanically steers its antennas to enhance cov erage and to provide signal enhancement to ward specic devices. Suc h capability enables improv ed cov erage uniformit y as w ell as high-rate directional links for mobile users in indo or scenarios. F rom the p erspective of electronically adjusting the an tenna b oresight, Huaw ei has pioneered “Smart An- tenna” solutions, which hav e b een widely deplo yed in the AirEngine Wi-Fi 6 APs and 5G BSs, as illustrated in Fig. 20(b) [113], [114]. Unlik e standard omnidirectional solutions, these systems typically emplo y a multi-elemen t smart an tenna array managed b y an intelligen t switch algorithm. This architecture enables selective activ ation of antenna elements to dynamically adjust the eective radiation direction of the array without mechanical mov e- men t. Such electronically-driven b oresight control enables fast adaptation to dynamic user mobility and c hannel v ariations. In summary , the protot yp es and commercial pro d- ucts review ed in this section collectively demonstrate the transition of RA technology from theoretical mod- eling to practical implementation. The diverse realization strategies review ed ab ov e highlight the practical v alue of exible antenna orientation/boresight control; how ever, they constitute merely a small subset of the p otential metho ds, suggesting considerable room for dev eloping new hardw are realization schemes. Lo oking ahead, antenna orien tation/b oresigh t control strategies are exp ected to ev olve b eyond single-mo dality sensing by incorp orating m ulti-mo dal sensor fusion. F urthermore, the incorporation of machine learning-based tra jectory prediction may help mitigate the latency incurred b y practical antenna b ore- sigh t adjustment and actuation. Finally , eorts should b e directed tow ard dev eloping cost-eective system ar- c hitectures and establishing general and standardized framew orks to facilitate the practical implementation and 29 (a) Antenna Element Area Boost Layout Boost Device T rack ing (a) TP-Link Archer AXE200 Omni (b) HUA WEI Smart Antenna Mechanically driven Electronically driven Fig. 20. Representativ e commercial products. (a) TP-Link’s Archer AXE200 Omni is a Wi-Fi 6E router equipp ed with four motorized antennas [111], [112]. (b) Hua wei’s smart an tennas adopt similar principles by dynamically switching among m ultiple radiating elements [113], [114]. widespread commercialization of this tec hnology . VI I. Extensions and F uture Directions In this section, w e discuss several representativ e exten- sions of RA-aided wireless systems and outline op en prob- lems that are critical for broadening their applications. A. Low-Altitude ISA C Lo w-altitude ISAC has emerged as a k ey enabler for LAE, demanding b oth high p erformance and deplo yment exibilit y to support real-time environmen tal p erception and reliable connectivity [115]. Ho wev er, conv entional lo w-altitude net works built up on xed-sector antenna infrastructures are primarily designed for 2D ground co v- erage with do wn tilt congurations optimized for terrestrial users. As a result, such arc hitectures are inherently limited in providing eective 3D spatial co verage and reliable trac king for dynamic aerial users op erating at v arying altitudes and trajectories. T o o vercome these limitations, the RA arc hitecture provides a exible hardware solu- tion that enhances spatial adaptability . By dynamically reconguring the b oresigh t direction, RA can eectively bridge cov erage gaps, improv e target tracking, and meet the stringent requiremen ts of LAE [116]. Within the ISAC paradigm, RAs can b e integrated into diverse deploymen t arc hitectures to simultaneously b o ost sensing precision and communication quality [71], [88], [89], [117]. A typical deplo yment is the ground-based RA architecture, where RAs are installed on terrestrial BSs to enhance co ver- age exibilit y [118]. This setup has the adv antage of a reliable p ow er supply and strong pro cessing capabilities, allo wing the system to p erform p ersisten t environmen tal scanning and high-precision beamforming. How ever, it also faces the dra wback of limited visibilit y in dense urban en vironments, where static obstacles frequently blo ck LoS links to lo w-altitude targets. T o tackle suc h issues, an alternativ e arc hitecture is the UA V-mounted RA (or aerial RA) system, which enables agile b oresight control in igh t [119]. In such systems, tra jectory planning and an tenna orientation are inheren tly coupled, and their join t adaptation can maintain or establish LoS connectivit y and supp ort sensing tasks during motion. Collectively , terres- trial and aerial RA deploymen ts provide complementary capabilities, enabling exible 3D cov erage and eective ISA C in dynamic LAE scenarios. Despite the enhanced spatial adaptability enabled by RA, realizing reliable lo w-altitude ISA C in practice re- mains challenging due to the highly dynamic nature of aerial platforms. In particular, UA V s typically op erate at high speeds with rapidly v arying tra jectories, whic h imp oses stringen t requiremen ts on real-time target track- ing and low-latency an tenna orientation adjustment. T o main tain reliable high-rate links and accurate sensing under such mobilit y , RA systems need to supp ort ag- ile boresight adjustmen t with fast actuation and high- precision control. F urthermore, robust and con tinuous connectivit y is critical for ensuring safe UA V operations, including real-time monitoring and regulatory sup ervision. Comm unication disruptions may lead to loss of control or degraded situational aw areness, thereby raising safety concerns in dynamic low-altitude airspace. Addressing these challenges is essen tial for realizing secure, reliable, and resilien t lo w-altitude ISA C systems. B. Cognitive Radio (CR) Systems CR systems enable secondary users (SUs) to access sp ectrum without causing harmful in terference to primary users (PUs), addressing the critical issue of sp ectrum scarcit y [120]. Due to the dynamic and spatially het- erogeneous distribution of sp ectrum resources, reliable sp ectrum sensing is essential to identify sp ectrum holes in b oth spatial and sp ectral domains [121]. How ever, traditional CR systems provide limited spatial adapt- abilit y , which makes it dicult to fully exploit suc h spatial sp ectrum opp ortunities. Integrating RA tec hnology in to CR systems introduces enhanced spatial exibilit y via adaptive b oresigh t control, enabling more exible and environmen t-aw are sp ectrum sensing and access. By 30 p erforming directional spatial scanning at the SU receiv er, an RA-enabled CR system can observe the radio environ- men t from multiple angular p ersp ectiv es and collect ne- grained sp ectrum-occupancy information. Once spatial sp ectrum holes are identied, the SU transceiver can supp ort opp ortunistic transmission by rotating its antenna b oresigh t to ward av ailable directions, thereby enhancing link eciency while limiting unintended interference to nearb y PUs. When sp ectrum resources are densely o ccu- pied and pure a voidance b ecomes insucient, RA-assisted CR systems can enable in terference-constrained sp ectrum sharing [86], [87]. In particular, joint optimization of transmit beamforming and antenna orien tation enables the SU to concentrate radiation p o wer tow ard intended receiv ers while spatially suppressing interference tow ard activ e PUs, thereby enhancing sp ectrum utilization in in terference-limited environmen ts. Despite these adv antages, several op en challenges re- main in realizing the full p oten tial of RA-aided CR systems. One critical issue lies in the practical uncertaint y or delay in obtaining accurate PU lo cation information, whic h ma y hinder precise boresight alignment and degrade sensing p erformance. Moreov er, p ersisten t sp ectrum mon- itoring is indisp ensable during dynamic sp ectrum access. In opp ortunistic access scenarios, once the spectrum is reo ccupied b y the PU, the RA system must rapidly recon- gure its an tenna orientations to ward alternativ e a v ailable directions or transition to sp ectrum-sharing access mo de with appropriate boresight control. This helps preserv e secondary link reliabilit y while constraining the induced in terference b elo w regulatory or prescrib ed thresholds. Ov erall, RA-aided CR systems present a promising di- rection to ward intelligen t spectrum sensing and access, and future eorts should emphasize the joint design of adaptiv e access strategies and dynamic b oresight con trol while enhancing robustness to PU uncertain ty . C. Physical La y er Security The growing demand for secure wireless communica- tions in complex and dynamic environmen ts has spurred increasing interest in physical lay er security (PLS) tech- niques [83]. By leveraging the inherent randomness and spatial characteristics of wireless channels, PLS pro- vides a complemen tary approac h to conv entional cryp- tographic metho ds, particularly in latency-sensitive or infrastructure-limited scenarios. How ever, conv entional PLS techniques typically rely on xed-antenna arc hi- tectures with limited spatial control to enhance the legitimate link while suppressing eav esdropp ers. With the integration of RA technology into wireless netw orks, exible antenna orien tation/b oresigh t con trol in tro duces additional spatial DoF s for secure transmission. In RA- enabled secure communication, antenna orientations can b e adaptiv ely adjusted to concentrate signal energy to- w ard the legitimate receiver while suppressing radiation to ward suspicious or vulnerable regions [83], [122]. This orien tation-induced spatial fo cusing strengthens the main c hannel and inherently reduces information leakage to p oten tial eav esdropp ers, thereby impro ving the achiev able secrecy rate without relying solely on p ow er-domain or signal-domain processing. Ho wev er, the same spatial adaptability ma y also p ose new securit y risks if adversarial no des are equipp ed with RA capabilities. F or example, an RA-enabled eav esdrop- p er can dynamically align its antenna b oresigh t to ward the legitimate transmitter to enhance its interception capabil- it y , while an RA-assisted jammer can directionally inject in terference into critical links with higher eciency and lo wer detectability . Suc h orientation-a ware attacks may signican tly degrade the secrecy performance of con ven- tional systems and c hallenge existing coun termeasures. T o address these emerging threats, adv anced RA-aided PLS strategies should incorp orate the join t design of antenna orien tation, transmission, and interference management. F or instance, an RA-equipp ed BS can proactively steer articial noise (AN), jamming signals, or deceptive w av e- forms tow ard suspicious spatial regions to suppress p o- ten tial ea vesdropping attempts, while preserving reliable transmission tow ard legitimate users. In addition, adaptive b oresigh t control combined with secure b eamforming can rapidly form spatial n ulls to w ard adv ersarial directions or recongure transmission paths up on threat detection. Ov erall, RA-aided PLS extends secrecy enhancement from signal-domain processing to spatial-domain recongura- tion. With prop er con trol and regulation, RA tec hnology oers a promising and recongurable hardw are-level solu- tion for next-generation secure wireless systems. D. Cell-F ree MIMO Net works Cell-free MIMO netw orks rely on a large n um b er of distributed APs to co op erativ ely serve users in a cell-less manner, oering macro-diversit y and eliminating inter- cell interference [123]. Ho wev er, their p erformance hinges hea vily on centralized co ordination and dense fron thaul signaling, p osing signicant challenges to training o ver- head, scalabilit y , and practical deploymen t. Integrating RA in to cell-free systems fundamentally alters this co- op eration paradigm. Instead of relying solely on digital preco ding to manage interference, eac h AP can reshape its eectiv e channel through antenna rotation, strengthening desired propagation paths while attenuating unfav orable directions b efore baseband pro cessing. Suc h orien tation- induced c hannel reshaping reduces link imbalance across users and allows multi-AP co operation to operate under more fav orable channel conditions. Recently , the authors in [81] rst incorp orated RA technology into cell-free net works, where each single-RA AP adjusts its antenna b oresigh t and p erforms AP-user pairing to enhance down- link p erformance. Extending b ey ond suc h pairing-based designs, the authors in [82] further inv estigated multi-RA arra y APs, where an tenna orientation and co op erativ e pre- co ding are jointly optimized, thereby revealing a substan- tially expanded design space. F urthermore, RA naturally promotes spatially selective co operation among APs. By 31 concen trating radiation p o w er and limiting co ordination to the most relev ant APs, RA systems reduce unnecessary co operation and signaling ov erhead. This facilitates more fron thaul-ecient compression and control strategies that are crucial for scalable deploymen ts, yielding a more fa vorable p erformance-complexit y trade-o. Nev ertheless, RA-aided cell-free MIMO systems also in tro duce distinct challenges. Since orien tation-dep enden t radiation gains are embedded into the eective c hannels, an tenna rotation becomes tightly coupled with co op erativ e preco ding and netw ork-level co ordination, making scal- able joint optimization non trivial. Moreov er, the AP-user asso ciation is no longer determined solely by pro ximity or large-scale fading, but it must consider the coupling b et w een discrete association decisions and antenna orien- tation con trol. F urthermore, coordinated b oresigh t control across a large n umber of APs remains dicult under limited fronthaul and imperfect CSI. F uture research ma y therefore in vestigate concepts for scalable b oresight con trol across distributed APs, joint AP-user asso ciation, and robust designs under imp erfect CSI and limited con trol signaling, thereby enabling practical RA-assisted cell-free deplo yments. E. Simultaneous Wireless Information and Po w er T ransfer (SWIPT) SWIPT has emerged as a promising solution for self- sustaining lo w-p o wer devices in future IoT and wireless sensor netw orks [124]. How ever, ac hieving an ecien t balance betw een information decoding and energy har- v esting remains a critical design c hallenge, esp ecially in dynamic and rich multipath wireless environmen ts [125]. The integration of RA technology unlocks new opp or- tunities for spatially adaptive transmission strategies, impro ving energy harv esting and information transfer eciency . By adaptively adjusting the antenna boresight direction, RAs can dynamically align with dominant c hannel paths or high-energy regions, thereb y improving the eective received p o wer for energy harv esting. In parallel, directional b oresigh t control enables selective signal reception to enhance the SINR for information deco ding. This exibility oers a new approach to manag- ing the fundamental SWIPT trade-o b eyond traditional b eamforming or pow er allocation techniques. F or instance, under fav orable channel conditions, the RA array can b e steered in an energy-dense direction to maximize harv ested p o wer, while in interference-limited regimes, it can prioritize directional alignmen t with information sources to ensure reliable comm unication. Moreov er, in the RA array , eac h antenna can be indep enden tly oriented to serv e dierent ob jectives. Some RAs may fo cus on maximizing SINR for data reception, while others align with regions of strong RF energy for harvesting. This spatial division of roles enables distributed optimization and functional decoupling, thereby enhancing the ov erall eciency of SWIPT netw orks. In practice, the gains of RA-aided SWIPT critically dep end on how accurately and how frequently antenna orien tations are up dated, since an tenna rotation incurs non-negligible actuation energy and latency . Therefore, the net energy eciency must account for b oth harv ested energy and an tenna rotation o verhead, which can shift the optimal rate-energy op erating p oin t, particularly for lo w-p o wer IoT devices. Beyond this o verhead issue, RA- aided SWIPT in tro duces a joint an tenna orien tation and resource allo cation problem across users with dierent service ob jectives. Although RA arrays enable spatial role division, dynamically coordinating antenna orientations to balance energy harvesting and information decoding ob jectiv es remains inherently complex. The problem is further complicated for co-lo cated SWIPT users, where the RF p o wer-maximizing direction do es not generally co- incide with the SINR-maximizing direction in the presence of interference. F uture researc h ma y therefore focus on ac hieving a fav orable trade-o b etw een antenna rotation o verhead and harv ested energy , developing RA orientation allo cation and scheduling for SWIPT, and theoretical c haracterization of the ac hiev able rate-energy region under practical rotation constraints. F. Other Miscellaneous T opics In addition to the ab o ve directions, RA tec hnology sho ws great p otential in a wide range of emerging applica- tions that require directional adaptability , spatial aw are- ness, and low-cost hardw are design. A unifying theme is that antenna b oresigh t/orientation control provides a light weigh t means of realizing directional alignmen t and spatial selectivit y at the radio (or transducer) fron t end, which is particularly relev an t to IMT-2030-oriented scenarios and highly dynamic, resource-constrained plat- forms [ 3 ]. Bey ond con ven tional wireless communication, the capa- bilit y of RA to realize lo w-complexity directional adapta- tion also makes it promising for a wide range of cross- disciplinary systems. F or instance, in optical comm unica- tion [126], RA enables dynamic alignment of highly direc- tional links, mitigating degradation caused by mobilit y , drift, or mec hanical oset. This is esp ecially benecial for aerial rela ys, space-ground free-space optical (FSO) systems, and high-precision visible ligh t comm unications. Similar b enets arise in acoustic communication [127], where rotatable acoustic transducers can adapt their b oresigh t in rich multipath or reectiv e environmen ts, suc h as underwater netw orks and smart audio systems, to impro ve signal reception and suppress interference without relying on large arra ys. The same directional exibility is also v aluable for sensing-related tasks. F or example, in radar, imaging, and en vironmental sensing [116], [128], [129], RA supports exible scanning and spatial selectivit y for high-resolution target detection on mobile or resource- constrained platforms. Its agilit y enables cost-eective p erception enhancement in UA V s, rob ots, and smart v ehicles. Moreo ver, for lo calization and navigation [130], [131], RA supports geometry-aided positioning and an- gular separation in dense multi-user settings. Last but 32 not least, RA can serve as a physical interface for AI- driv en wireless systems. The antenna b oresigh t direction can b e optimized using reinforcement learning or predic- tiv e mo dels, enabling autonomous and environmen t-aw are b oresigh t control with minimal signaling ov erhead. When in tegrated with edge in telligence, RA empow ers real-time spatial decision-making, supp orting self-organizing and adaptiv e netw ork op erations. VI II. Conclusions By enabling exible 3D antenna orientation/boresight rotation, RA in tro duces additional spatial DoF s that enhance wireless communication and sensing p erformance without requiring extra antenna resources or additional deplo yment space. Compared with conv entional xed- an tenna architectures, RA oers a ligh tw eight and prac- tical means of spatial adaptation for future wireless net works. In particular, RA enables directional “spotlight” transmission tow ard desired users or targets and eye-lik e spatial scanning of the surrounding en vironment, thereby op ening new p ossibilities for communication and sensing system designs. In this pap er, w e hav e pro vided a com- prehensiv e tutorial on RA-emp ow ered wireless netw orks. T o this end, we hav e reviewed the historical developmen t of RA-related technologies, established a unied frame- w ork for antenna/arra y rotation and channel mo deling, and inv estigated rotation optimization i n represen tative comm unication and ISA C scenarios. In addition, we hav e discussed RA channel estimation/acquisition, practical hardw are arc hitectures, deplo yment trade-os, and recent protot yp es that demonstrate the feasibility and poten- tial of RA-enabled systems. Overall, RA strikes a com- p elling balance b etw een spatial exibility , implementation complexit y , and deploymen t cost, oering a promising path wa y to ward adaptive cov erage, reliable transmission in interference-limited en vironments, and high-resolution sensing in emerging 6G systems. Lo oking ahead, further adv ances in the co-design of rotation optimization, c hannel estimation/acquisition, and practical RA implementation will contin ue to unlo c k the p oten tial of RA for more agile, ecient, and intelligen t wireless comm unication and sensing. References [1] M. Z. Chowdh ury , M. Shahjalal, S. Ahmed, and Y. M. Jang, “6G wireless comm unication systems: Applications, require- ments, technologies, challenges, and researc h directions,” IEEE Open J. Commun. So c., vol. 1, pp. 957–975, Jul. 2020. [2] I. F. Akyildiz, A. Kak, and S. Nie, “6G and beyond: The future of wireless communications systems,” IEEE Access, vol. 8, pp. 133 995–134 030, Jul. 2020. [3] F ramework and Overall Ob jectives of the F uture Developmen t of IMT for 2030 and Beyond, International T elecommunication Union Rec. ITU-R M.2160-0, Geneva, Switzerland, Nov. 2023. [4] C.-X. W ang, X. Y ou, X. Gao, X. Zh u, Z. Li, C. Zhang, H. W ang, Y. Huang, Y. Chen, H. Haas, J. S. Thompson, E. G. Larsson, M. D. Renzo, W. T ong, P . Zhu, X. Shen, H. V. Poor, and L. Hanzo, “On the road to 6G: Visions, requirements, k ey technologies, and testb eds,” IEEE Commun. Surveys T uts., vol. 25, no. 2, pp. 905–974, 2nd Quart. 2023. [5] X. W ang, J. Mei, S. Cui, C.-X. W ang, and X. S. Shen, “Re- alizing 6G: The op erational goals, enabling technologies of fu- ture net works, and v alue-oriented intelligen t multi-dimensional multiple access,” IEEE Netw., vol. 37, no. 1, pp. 10–17, Jan. 2023. [6] C. Y ou, Y. Cai, Y. Liu, M. Di Renzo, T. M. Duman, A. Y ener, and A. L. Swindleh urst, “Next generation adv anced transceiv er technologies for 6G and beyond,” IEEE J. Sel. Areas Commun., vol. 43, no. 3, pp. 582–627, Mar. 2025. [7] F. T ariq, M. R. Khandaker, K.-K. W ong, M. A. Imran, M. Bennis, and M. Debbah, “A sp eculativ e study on 6G,” IEEE Wireless Commun., vol. 27, no. 4, pp. 118–125, Aug. 2020. [8] W. Saad, M. Bennis, and M. Chen, “A vision of 6G wireless systems: Applications, trends, technologies, and open researc h problems,” IEEE Netw., vol. 34, no. 3, pp. 134–142, Ma y/Jun. 2020. [9] A. J. Paulraj, D. A. Gore, R. U. Nabar, and H. Bolcskei, “An ov erview of MIMO communications - a k ey to gigabit wireless,” Proc. IEEE, vol. 92, no. 2, pp. 198–218, F eb. 2004. [10] L. Lu, G. Y. Li, A. L. Swindlehurst, A. Ashikhmin, and R. Zhang, “An ov erview of massive MIMO: Benets and challenges,” IEEE J. Sel. T opics Signal Pro cess., vol. 8, no. 5, pp. 742–758, Oct. 2014. [11] E. G. Larsson, O. Edfors, F. T ufvesson, and T. L. Marzetta, “Massive MIMO for next generation wireless systems,” IEEE Commun. Mag., vol. 52, no. 2, pp. 186–195, F eb. 2014. [12] H. Lu, Y. Zeng, C. Y ou, Y. Han, J. Zhang, Z. W ang, Z. Dong, S. Jin, C.-X. W ang, T. Jiang, X. Y ou, and R. Zhang, “A tutorial on near-eld XL-MIMO communications to ward 6G,” IEEE Commun. Surv eys T uts., vol. 26, no. 4, pp. 2213–2257, 4th Quart. 2024. [13] Z. W ang, J. Zhang, H. Du, D. Niy ato, S. Cui, B. Ai, M. Debbah, K. B. Letaief, and H. V. Poor, “A tutorial on extremely large- scale MIMO for 6G: F undamen tals, signal pro cessing, and applications,” IEEE Commun. Surv eys T uts., vol. 26, no. 3, pp. 1560–1605, 3rd Quart. 2024. [14] J. Hoydis, S. ten Brink, and M. Debbah, “Massiv e MIMO: How many antennas do we need?” in Pro c. IEEE 49th Ann. Allerton Conf. on Comm. Control, and Computing (Allerton), Monticello, IL, USA, Sep. 2011, pp. 545–550. [15] H. W ang and Y. Zeng, “Can sparse arra ys outp erform col- located arra ys for future wireless comm unications?” in Pro c. IEEE Global Commun. Conf. (GLOBECOM) W orkshops, Kuala Lumpur, Mala ysia, Dec. 2023, pp. 667–672. [16] Y. Zeng and R. Zhang, “Millimeter wa ve MIMO with lens antenna array: A new path division multiplexing paradigm,” IEEE T rans. Commun., v ol. 64, no. 4, pp. 1557–1571, Apr. 2016. [17] B. Zheng, C. Y ou, W. Mei, and R. Zhang, “A survey on chan- nel estimation and practical passiv e beamforming design for intelligen t reecting surface aided wireless comm unications,” IEEE Commun. Surveys T uts., vol. 24, no. 2, pp. 1035–1071, 2nd Quart. 2022. [18] Q. W u, S. Zhang, B. Zheng, C. Y ou, and R. Zhang, “Intelligent reecting surface-aided wireless communications: A tutorial,” IEEE T rans. Commun., vol. 69, no. 5, pp. 3313–3351, May 2021. [19] B. Zheng and R. Zhang, “Intelligen t reecting surface- enhanced OFDM: Channel estimation and reection optimiza- tion,” IEEE Wireless Commun. Lett., v ol. 9, no. 4, pp. 518–522, Apr. 2020. [20] C. Huang, A. Zapp one, G. C. Alexandrop oulos, M. Debbah, and C. Y uen, “Recongurable intelligen t surfaces for energy eciency in wireless communication,” IEEE T rans. Wireless Commun., vol. 18, no. 8, pp. 4157–4170, Aug. 2019. [21] C. Zhou, C. Y ou, C. Zhou, L. Y ao, W. Y uan, B. Zheng, and N. W u, “Rotatable IRS-assisted 6DMA communications: A tw o-timescale design,” arXiv preprint arXiv:2512.15092, 2025. [22] Q. W u, B. Zheng, C. Y ou, L. Zhu, K. Shen, X. Shao, W. Mei, B. Di, H. Zhang, E. Basar, L. Song, M. D. Renzo, Z.-Q. Luo, and R. Zhang, “In telligent surfaces empow ered wireless netw ork: Recent adv ances and the road to 6G,” Pro c. IEEE, vol. 112, no. 7, pp. 724–763, Jul. 2024. [23] B. Zheng, Q. W u, T. Ma, and R. Zhang, “Rotatable antenna enabled wireless communication: Mo deling and optimization,” IEEE T rans. Commun., 2026, Early Access. 33 [24] Q. W u, B. Zheng, T. Ma, and R. Zhang, “Mo deling and optimization for rotatable antenna enabled wireless communi- cation,” in Pro c. IEEE Int. Conf. Commun. (ICC), Montreal, Canada, Jun. 2025, pp. 1–6. [25] B. Zheng, T. Ma, C. Y ou, J. T ang, R. Schober, and R. Zhang, “Rotatable antenna enabled wireless comm unication and sens- ing: Opportunities and challenges,” IEEE Wireless Commun., Oct. 2025, Early Access. [26] X. Xiong, B. Zheng, W. W u, W. Zhu, M. W en, S. Lin, and Y. Zeng, “Intelligen t rotatable antenna for integrated sensing, communication, and computation: Challenges and opportunities,” IEEE Wireless Comm un., vol. 33, no. 1, pp. 173–180, F eb. 2026. [27] C.-W. Baek, S. Song, J.-H. Park, S. Lee, J.-M. Kim, W. Choi, C. Cheon, Y.-K. Kim, and Y. Kw on, “A V-band microma- chined 2-D b eam-steering an tenna driven by magnetic force with p olymer-based hinges,” IEEE T rans. Microw. Theory T echn., vol. 51, no. 1, pp. 325–331, Jan. 2003. [28] T. Y ang, Y. Chen, X. Li, K. Zhang, J. He, and W. Su, “A low-pow er piezo electric MEMS motor using standing w av e,” Sens. Actuators A: Phys., vol. 387, p. 116367, Jun. 2025. [29] Y. Zhang, Z. Han, S. T ang, S. Shen, C.-Y. Chiu, and R. Murc h, “A highly pattern-recongurable planar antenna with 360 ◦ single- and multi-beam steering,” IEEE T rans. Antennas Propag., vol. 70, no. 8, pp. 6490–6504, Aug. 2022. [30] M. A. T o wq, I. Bahceci, S. Blanch, J. Romeu, L. Jofre, and B. A. Cetiner, “A recongurable antenna with beam steering and beamwidth v ariability for wireless comm unications,” IEEE T rans. Antennas and Propag., vol. 66, no. 10, pp. 5052–5063, Oct. 2018. [31] P . Lot, S. Soltani, and R. D. Murc h, “Printed endre b eam- steerable pixel antenna,” IEEE T rans. Antennas and Propag., vol. 65, no. 8, pp. 3913–3923, Aug. 2017. [32] W. Ma, L. Zhu, Y. T an, B. Zheng, Y. Zhang, Y. Zhang, K. Ying, Z. Gao, H. Sun, X. Shao, Z. Xiao, D. Niy ato, and R. Zhang, “A survey on recongurable and mov able antennas for wireless communications and sensing,” IEEE Comm un. Surveys T uts., vol. 28, pp. 4842–4882, F eb. 2026. [33] G. J. Holzmann and B. Pehrson, The Early History of Data Netw orks. W ashington, DC, USA: IEEE Computer So ciety Press, 1995. [34] C. A. Balanis, Antenna Theory: Analysis and Design. John Wiley & Sons, 2015. [35] U.S. Navy Bureau of Ships, Instructions for Radio T ransmit- ting Equipmen t Mo del YE-1, Navy Department, W ashington, DC, USA, 1944, technical Manual for shipboard directional homing b eacon. [36] I. Getting, “SCR-584 radar and the Mark 56 nav al gun re control system,” IEEE Aerosp. Electron. Syst. Mag., vol. 5, no. 10, pp. 3–15, Oct. 1990. [37] D. Chauvel, N. Haese, P .-A. Rolland, D. Collard, and H. F ujita, “A micro-machined microw av e antenna integrated with its electrostatic spatial scanning,” in Pro c. IEEE Int. MEMS W orkshop, Nagoy a, Japan, Jan. 1997, pp. 84–89. [38] J.-C. Chiao, Y. F u, I. M. Chio, M. DeLisio, and L.-Y. Lin, “MEMS recongurable V ee antenna,” in IEEE MTT-S Int. Microw av e Symp. Dig., Anaheim, CA, Jun. 1999, pp. 1515– 1518. [39] R. Harrington, “Reactively controlled directive arrays,” IEEE T rans. An tennas Propag., v ol. 26, no. 3, pp. 390–395, May 1978. [40] D. V. Thiel and S. Smith, Switched Parasitic An tennas for Cellular Communications. Artec h House, 2002. [41] S. Nik olaou, R. Baira v asubramanian, C. Lugo, I. Car- rasquillo, D. Thompson, G. Ponc hak, J. Papapolymerou, and M. T entzeris, “Pattern and frequency recongurable annu- lar slot antenna using PIN dio des,” IEEE T rans. An tennas Propag., vol. 54, no. 2, pp. 439–448, F eb. 2006. [42] L. Zhu and K.-K. W ong, “Historical review of uid antenna and mov able antenna,” arXiv preprin t arXiv:2401.02362, 2024. [43] C. A. Balanis, Mo dern An tenna Handb ook. Hob oken, New Jersey , USA: John Wiley & Sons, 2007. [44] S. Zhao, H. Y ang, and H. Y ang, “Single antenna spatial diversit y ,” in Pro c. IEEE Int. Conf. Wireless Comm un., Netw., Mobile Comput. (WiCOM), Beijing, China, Sep. 2009, pp. 1–4. [45] D. W. S. T am, “Electrolytic uid antenna,” US Paten t US7898484B1, 2008. [46] Y. K osta and B. Chaturvedi, “A liquid patc h microw av e antenna,” Pro c. NACONECS-1989, Dept. Electron. Comput. Eng., Univ. Ro orkee, India, T ata McGraw-Hill, pp. 43–47, 1989. [47] Y. Kosta and S. Kosta, “Liquid antenna systems,” in Pro c. IEEE An tennas Propag. So c. Symp., Monterey , CA, USA, Jun. 2004, pp. 2392–2395. [48] K.-K. W ong, A. Sho jaeifard, K.-F. T ong, and Y. Zhang, “Performance limits of uid an tenna systems,” IEEE Comm un. Lett., vol. 24, no. 11, pp. 2469–2472, Nov. 2020. [49] K.-K. W ong, K.-F. T ong, Y. Zhang, and Z. Zhongbin, “Fluid antenna system for 6G: When Bruce Lee inspires wireless communications,” Electron. Lett., v ol. 56, no. 24, pp. 1288– 1290, Nov. 2020. [50] K.-K. W ong, A. Sho jaeifard, K.-F. T ong, and Y. Zhang, “Fluid antenna systems,” IEEE T rans. Wireless Commun., v ol. 20, no. 3, pp. 1950–1962, Mar. 2021. [51] W. K. New, K.-K. W ong, H. Xu, C. W ang, F. R. Ghadi, J. Zhang, J. Rao, R. Murch, P . Ramírez-Espinosa, D. Morales- Jimenez, C.-B. Chae, and K.-F. T ong, “A tutorial on uid an- tenna system for 6G netw orks: Encompassing communication theory , optimization metho ds and hardware designs,” IEEE Commun. Surveys T uts., vol. 27, no. 4, pp. 2325–2377, Aug. 2025. [52] W.-J. Lu, C.-X. He, Y. Zhu, K.-F. T ong, K.-K. W ong, H. Shin, and T. J. Cui, “Fluid antennas: Reshaping intrinsic properties for exible radiation c haracteristics in intelligen t wireless net- works,” IEEE Commun. Mag., vol. 63, no. 5, pp. 40–45, May 2025. [53] W. K. New, K.-K. W ong, C. W ang, C.-B. Chae, R. Murch, H. Jafarkhani, and Y. Hao, “Fluid an tenna systems: Redening recongurable wireless comm unications,” IEEE J. Sel. Areas Commun., vol. 44, pp. 1013–1044, Nov. 2025. [54] H. Hong, K.-K. W ong, C.-B. Chae, H. Xu, X. Guo, F. R. Ghadi, Y. Chen, Y. Xu, B. Liu, K.-F. T ong, and Y. Zhang, “A contemporary survey on uid antenna systems: F undamentals and netw orking persp ectiv es,” IEEE T rans. Net w. Sci. Eng., vol. 13, pp. 2305–2328, Sep. 2025. [55] T. W u, K. Zhi, J. Y ao, X. Lai, J. Zheng, H. Niu, M. Elkashlan, K.-K. W ong, C.-B. Chae, Z. Ding, G. K. Karagiannidis, M. Debbah, and C. Y uen, “Fluid antenna systems enabling 6G: Principles, applications, and research directions,” IEEE Wireless Commun., 2025, Early Access. [56] L. Zh u, W. Ma, and R. Zhang, “Modeling and performance analysis for mov able antenna enabled wireless communica- tions,” IEEE T rans. Wireless Commun., vol. 23, no. 6, pp. 6234–6250, Jun. 2024. [57] W. Ma, L. Zhu, and R. Zhang, “MIMO capacity characteri- zation for mov able antenna systems,” IEEE T rans. Wireless Commun., vol. 23, no. 4, pp. 3392–3407, Apr. 2024. [58] L. Zhu, W. Ma, B. Ning, and R. Zhang, “Mov able-antenna enhanced multiuser communication via antenna p osition opti- mization,” IEEE T rans. Wireless Commun., vol. 23, no. 7, pp. 7214–7229, Jul. 2024. [59] X. Shao, Q. Jiang, and R. Zhang, “6D mov able antenna based on user distribution: Mo deling and optimization,” IEEE T rans. Wireless Commun., vol. 24, no. 1, pp. 355–370, Jan. 2025. [60] X. Shao, R. Zhang, Q. Jiang, and R. Schober, “6D mov able antenna enhanced wireless netw ork via discrete p osition and rotation optimization,” IEEE J. Sel. Areas Commun., vol. 43, no. 3, pp. 674–687, Mar. 2025. [61] X. Shao and R. Zhang, “6DMA enhanced wireless netw ork with exible antenna p osition and rotation: Opportunities and challenges,” IEEE Commun. Mag., vol. 63, no. 4, pp. 121–128, Apr. 2025. [62] A. F ukuda, H. Y amamoto, H. Okazaki, Y. Suzuki, and K. Kaw ai, “Pinching antenna: Using a dielectric wa veguide as an antenna,” NTT DOCOMO T echnical J., v ol. 23, no. 3, pp. 5–12, Jan. 2022. [63] K.-K. W ong, K.-F. T ong, Z. Chu, and Y. Zhang, “A vision to smart radio en vironment: Surface wa ve comm unication superhighwa ys,” IEEE Wireless Commun., vol. 28, no. 1, pp. 112–119, F eb. 2020. [64] Z. Ding, R. Schober, and H. V. Poor, “Flexible-antenna systems: A pinching-an tenna p erspective,” IEEE T rans. Com- mun., vol. 73, no. 10, pp. 9236–9253, Oct. 2025. 34 [65] Y. Liu, Z. W ang, X. Mu, C. Ouyang, X. Xu, and Z. Ding, “Pinching-an tenna systems: Architecture designs, opp ortuni- ties, and outlo ok,” IEEE Commun. Mag., vol. 64, no. 1, pp. 190–196, Jan. 2026. [66] L. Zh u, W. Ma, and R. Zhang, “Mov able antennas for wireless communication: Opp ortunities and challenges,” IEEE Com- mun. Mag., vol. 62, no. 6, pp. 114–120, Jun. 2024. [67] L. Zhu, W. Ma, W. Mei, Y. Zeng, Q. W u, B. Ning, Z. Xiao, X. Shao, J. Zhang, and R. Zhang, “A tutorial on mov able antennas for wireless netw orks,” IEEE Commun. Surv eys T uts., vol. 28, pp. 3002–3054, F eb. 2026. [68] L. Zhu, W. Ma, and R. Zhang, “Mov able-antenna array enhanced b eamforming: A chieving full arra y gain with null steering,” IEEE Commun. Lett., vol. 27, no. 12, pp. 3340– 3344, Dec. 2023. [69] W. Ma, L. Zhu, and R. Zhang, “Compressed sensing based channel estimation for mov able antenna communications,” IEEE Commun. Lett., v ol. 27, no. 10, pp. 2747–2751, Oct. 2023. [70] X. Shao, W. Mei, C. Y ou, Q. W u, B. Zheng, C.-X. W ang, J. Li, R. Zhang, R. Schober, L. Zh u, W. Zhuang, and X. Shen, “A tutorial on six-dimensional mov able antenna for 6G netw orks: Synergizing p ositionable and rotatable antennas,” IEEE Com- mun. Surveys T uts., vol. 28, pp. 3666–3709, Aug. 2025. [71] S. Li, J. T ang, B. Zheng, L. Zhu, C. Y ang, N. Zhao, X. Y. Zhang, and K.-K. W ong, “Rotatable antenna system em- powered lo w-altitude economy: Opportunities and challenges,” IEEE Wireless Comm un., vol. 33, no. 1, pp. 116–123, F eb. 2025. [72] H. Goldstein, C. Poole, and J. Safko, Classical Mechanics (Third Edition). Addison-W esley , San F rancisco, 2001. [73] J. Dieb el, “Representing attitude: Euler angles, unit quater- nions, and rotation vectors,” Matrix, vol. 58, no. 15-16, pp. 1–35, 2006. [74] 3GPP TR 38.901, “5G; Study on c hannel model for frequencies from 0.5 to 100 GHz,” do cumen t 3GPP TR 38.901 version 16.1.0 Release 16, 2020. [75] X. Shao, R. Zhang, Q. Jiang, J. Park, T. Q. S. Quek, and R. Schober, “Distributed channel estimation and optimization for 6D mov able antenna: Unv eiling directional sparsity ,” IEEE J. Sel. T op. Signal Pro cess., vol. 19, no. 2, pp. 349–365, Mar. 2025. [76] F. L. Markley and J. L. Crassidis, F undamentals of Spacecraft Attitude Determination and Control. New Y ork, NY, USA: Springer, 2014. [77] M. Rebato, J. Park, P . P op ovski, E. De Carv alho, and M. Zorzi, “Stochastic geometric co verage analysis in mm wa ve cellular netw orks with realistic channel and antenna radiation mo dels,” IEEE T rans. Commun., vol. 67, no. 5, pp. 3736–3752, 2019. [78] R. Zhang, Y. Shao, and Y. C. Eldar, “P olarization-aw are mov able antenna,” IEEE T rans. Wireless Commun., vol. 25, pp. 7428–7442, Nov. 2025. [79] C. Zhang, H. Zhou, R. Long, Y.-C. Liang, and B.-H. Soong, “Polarization-a ware recongurable antenna aided wire- less communications,” arXiv preprint arXiv:2603.01166, Mar. 2026. [80] J. Ding, Z. Zhou, X. Shao, B. Jiao, and R. Zhang, “Polar- forming for wireless net works: Opp ortunities and challenges,” IEEE Commun. Mag., vol. 64, no. 3, pp. 118–124, Mar. 2026. [81] K. Pan, B. Zheng, Y. T an, E. Björnson, R. Schober, and R. Zhang, “Rotatable antenna-enhanced cell-free communica- tion,” arXiv preprint arXiv:2512.04742, 2025. [82] X. Peng, Q. W u, Z. Zheng, Y. Zhu, W. Chen, P . Huang, Y. Gao, and H. W ang, “Cell-free MIMO with rotatable antennas: When macro-diversit y meets antenna directivity ,” arXiv preprint arXiv:2601.16543, 2026. [83] L. Dai, B. Zheng, Q. W u, C. Y ou, R. Schober, and R. Zhang, “Rotatable antenna-enabled secure wireless communication,” IEEE Wireless Commun. Lett., vol. 14, no. 11, pp. 3440–3444, Nov. 2025. [84] X. Xiong, B. Zheng, W. W u, X. Shao, L. Dai, M.-M. Zhao, and J. T ang, “Ecient channel estimation for rotatable antenna- enabled wireless communication,” IEEE Wireless Commun. Lett., vol. 14, no. 11, pp. 3719–3723, Nov. 2025. [85] C. Zhou, C. Y ou, B. Zheng, X. Shao, and R. Zhang, “Rotatable antennas for in tegrated sensing and communications,” IEEE Wireless Commun. Lett., vol. 14, no. 9, pp. 2838–2842, Sep. 2025. [86] X. P eng, Q. W u, Z. Zheng, W. Chen, Y. Zhu, and Y. Gao, “Rotatable antenna enabled sp ectrum sharing: Joint an- tenna orientation and beamforming design,” arXiv preprint arXiv:2509.19912, 2025. [87] Y. T an, B. Zheng, Y. F ang, D. W. Kwan Ng, J. Xu, and R. Zhang, “Rotatable an tenna-enabled sp ectrum sharing in cognitive radio systems,” IEEE Wireless Commun. Lett., vol. 15, pp. 1732–1736, Jan. 2026. [88] Q. W u, B. Zheng, C. Y ou, and J. T ang, “Rotatable an tenna enhanced integrated sensing and communication,” in Proc. IEEE GLOBECOM W orkshops (GC Wkshps), T aip ei, T aiwan, Dec. 2025, pp. 1–6. [89] Y. Zhang, H. C. So, D. Niyato, and C. Masouros, “Rotatable antennas for near-eld integrated sensing and communication,” IEEE T rans. Wireless Commun., vol. 25, pp. 10 986–11 001, Jan. 2026. [90] Q. W ang, B. Zheng, X. Xiong, W. Mei, C. Y ou, Q. W u, and J. T ang, “Rotatable an tenna-enabled mobile edge computing,” arXiv preprint arXiv:2603.16275, 2026. [91] S. Ra jan, S. W ang, R. Inkol, and A. Joyal, “Ecient approx- imations for the arctangent function,” IEEE Signal Process. Mag., vol. 23, no. 3, pp. 108–111, May 2006. [92] R. Schmidt, “Multiple emitter location and signal parameter estimation,” IEEE T rans. Antennas Propag., v ol. 34, no. 3, pp. 276–280, Mar. 1986. [93] J. Lee, G.-T. Gil, and Y. H. Lee, “Channel estimation via orthogonal matching pursuit for hybrid MIMO systems in millimeter wa ve communications,” IEEE T rans. Commun., vol. 64, no. 6, pp. 2370–2386, Jun. 2016. [94] M. A. Dav enport, D. Needell, and M. B. W akin, “Signal space CoSaMP for sparse recov ery with redundant dictionaries,” IEEE T rans. Inf. Theory , vol. 59, no. 10, pp. 6820–6829, Oct. 2013. [95] N. Nguyen, D. Needell, and T. W oolf, “Linear conv ergence of stochastic iterativ e greedy algorithms with sparse constrain ts,” IEEE T rans. Inf. Theory , vol. 63, no. 11, pp. 6869–6895, Nov. 2017. [96] R. Zhang, L. Cheng, S. W ang, Y. Lou, W. W u, and D. W. K. Ng, “T ensor decomp osition-based channel estimation for hy- brid mm wa ve massiv e MIMO in high-mobilit y scenarios,” IEEE T rans. Commun., vol. 70, no. 9, pp. 6325–6340, Sep. 2022. [97] P . Dong, H. Zhang, G. Y. Li, I. S. Gaspar, and N. Naderi- Alizadeh, “Deep CNN-based channel estimation for mmw av e massive MIMO systems,” IEEE J. Sel. T opics Signal Pro cess., vol. 13, no. 5, pp. 989–1000, Sep. 2019. [98] J. Suh, C. Kim, W. Sung, J. So, and S. W. Heo, “Construction of a generalized DFT co debo ok using channel-adaptiv e param- eters,” IEEE Commun. Lett., vol. 21, no. 1, pp. 196–199, Jan. 2017. [99] M. H. Mohd No or and A. O. Ige, “A survey on state-of-the-art deep learning applications and challenges,” Eng. Appl. Artif. Intell., vol. 159, pp. 1–48, Jul. 2025. [100] W. Y. B. Lim, N. C. Luong, D. T. Hoang, Y. Jiao, Y.-C. Liang, Q. Y ang, D. Niyato, and C. Miao, “F ederated learning in mobile edge net works: A comprehensiv e survey ,” IEEE Commun. Surveys T uts., vol. 22, no. 3, pp. 2031–2063, Apr. 2020. [101] N. Jia, Z. Qu, B. Y e, Y. W ang, S. Hu, and S. Guo, “A compre- hensive survey on communication-ecient federated learning in mobile edge environmen ts,” IEEE Commun. Surveys T uts., vol. 27, no. 6, pp. 3710–3741, Dec. 2025. [102] L. Dai, B. Zheng, Y. T an, L. Zhu, F. Chen, and R. Zhang, “Rotatable antenna enabled wireless communication system with visual recognition: A prototype implemen tation,” in Pro c. IEEE/CIC Int. Conf. Comm un. China (ICCC), Shanghai, China, Aug. 2025, pp. 1–6. [103] Q. Dai, B. Zheng, Q. W ang, X. Xiong, X. Shao, L. Zhu, and R. Zhang, “A demo of radar sensing aided rotatable antenna for wireless communication system,” in Pro c. IEEE Int. Conf. Electron. Circuits Inf. Eng. (ECIE), Guangzhou, China, May 2025, pp. 205–208. [104] K. M. Morshed, D. K. Karmokar, K. P . Esselle, and L. Mateko vits, “Beam-switching antennas for 5G millimeter- wa ve wireless terminals,” Sensors, vol. 23, no. 14, pp. 6285– 6305, Jul. 2023. 35 [105] M. Boy arsky , T. Sleasman, M. F. Imani, J. N. Gollub, and D. R. Smith, “Electronically steered metasurface an tenna,” Sci. Rep., vol. 11, no. 1, pp. 1–10, F eb. 2021. [106] J. R. Lindle, A. T. W atnik, and V. A. Cassella, “Ecient multibeam large-angle nonmechanical laser beam steering from computer-generated holograms rendered on a liquid crystal spatial light mo dulator,” Appl. Opt., vol. 55, no. 16, pp. 4336– 4341, Jun. 2016. [107] X. Li, H. Min, Y. Zeng, S. Jin, L. Dai, Y. Y uan, and R. Zhang, “Sparse MIMO for ISAC: New opportunities and challenges,” IEEE Wireless Commun., vol. 32, no. 4, pp. 170–178, Aug. 2025. [108] M. R. Hay al, E. E. Elsay ed, D. Kakati, M. Singh, A. Elkky , A. I. Boghdady , A. Grover, S. Mehta, S. A. H. Mohsan, and I. Nurhiday at, “Mo deling and inv estigation on the p er- formance enhancemen t of ho vering UA V‑based FSO relay optical wireless communication systems under p ointing errors and atmospheric turbulence eects,” Opt. Quantum Electron., vol. 55, no. 625, pp. 1–23, Jul. 2023. [109] J. Redmon, S. Divv ala, R. Girshick, and A. F arhadi, “Y ou only look once: Unied, real-time ob ject detection,” in Pro c. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Las V egas, NV, USA, Jun. 2016, pp. 779–788. [110] Y. Du, Z. Zhao, Y. Song, Y. Zhao, F. Su, T. Gong, and H. Meng, “Strongsort: Mak e deepsort great again,” IEEE T rans. Multimedia, vol. 25, pp. 8725–8737, 2023. [111] TP-Link Systems Inc., Archer AXE200 Omni, Accessed: Dec. 21, 2025. [Online], A v ailable: https://www.tp- link.com/en/ home- networking/wifi- router/archer- axe200- omni/. [112] TP-Link Systems Inc., AXE11000 T ri-Band Wi-Fi 6E Motor Router, Accessed: Dec. 21, 2025. [Online], A v ailable: https://static.tp- link.com/upload/product- overview/2023/ 202304/20230428/Archer%20AXE200%20Omni(US)1.0_ Datasheet.pdf. [113] Hua wei T echnologies Co., Ltd., What Are Smart Antennas?, Accessed: Dec. 21, 2025. [Online], A v ailable: https://info.support.huawei.com/info- finder/encyclopedia/ en/Smart+Antenna.html. [114] Hua wei T echnologies Co., Ltd., Wi-Fi Smart Antenna T ech- nology , Accessed: Dec. 21, 2025. [Online], A v ailable: https: //support.huawei.com/enterprise/en/doc/EDOC1100221301. [115] Y. Song, Y. Zeng, Y. Y ang, Z. Ren, G. Cheng, X. Xu, J. Xu, S. Jin, and R. Zhang, “An overview of cellular ISAC for low-altitude UA V: New opp ortunities and c hallenges,” IEEE Commun. Mag., vol. 63, no. 12, pp. 88–95, Dec. 2025. [116] J. Jiang, F. Shu, B. Deng, M. Li, J. Bai, Y. W ang, C. P an, and J. W ang, “Rotatable antenna-arra y-enhanced direction- sensing for lo w-altitude communication netw ork: Method and performance,” arXiv preprint arXiv:2512.00435, 2025. [117] Z. W ang, L. Yin, Z. Lei, Y. Liu, Y. Sun, and H. Y ang, “Rotatable arra y-aided h ybrid beamforming for integrated sensing and comm unication,” IEEE Internet Things J., Dec. 2025, Early Access. [118] Z. Zheng, Q. W u, Y. Zh u, H. W ang, Y. Gao, W. Chen, and J. Xiong, “Lo w-altitude ISA C with rotatable activ e and passive arrays,” arXiv preprint arXiv:2512.20987, 2025. [119] S. Chen, G. Chen, L. Shi, Q. W u, and K. W ei, “Rotatable an- tenna meets UA V: T ow ards dual-lev el channel reconguration paradigm for ISAC,” arXiv preprint arXiv:2510.15295, 2025. [120] R. Zhang, Y.-C. Liang, and S. Cui, “Dynamic resource allo ca- tion in cognitive radio netw orks,” IEEE Signal Pro cess. Mag., vol. 27, no. 3, pp. 102–114, May 2010. [121] R. Saruthirathanaw orakun, J. M. Peha, and L. M. Correia, “Opportunistic sharing b et ween rotating radar and cellular,” IEEE J. Sel. Areas Commun., vol. 30, no. 10, pp. 1900–1910, Nov. 2012. [122] P . Jiang, Q. Li, L. Mai, and Q. Zhang, “A verage secrecy capacity maximization of rotatable antenna-assisted secure communications,” arXiv preprint arXiv:2511.18097, 2025. [123] H. A. Ammar, R. Adv e, S. Shahbazpanahi, G. Boudreau, and K. V. Srinivas, “User-centric cell-free massive MIMO netw orks: A survey of opp ortunities, challenges and solutions,” IEEE Commun. Surveys T uts., v ol. 24, no. 1, pp. 611–652, 1st Quart. 2021. [124] T. D. P . P erera, D. N. K. Jay akody , S. K. Sharma, S. Chatzino- tas, and J. Li, “Simultaneous wireless information and p o wer transfer (SWIPT): Recent adv ances and future c hallenges,” IEEE Commun. Surveys T uts., v ol. 20, no. 1, pp. 264–302, 1st Quart. 2017. [125] G. Pan, H. Lei, Y. Y uan, and Z. Ding, “Performance analysis and optimization for SWIPT wireless sensor netw orks,” IEEE T rans. Commun., vol. 65, no. 5, pp. 2291–2302, May 2017. [126] H. Kaushal and G. Kaddoum, “Optical communication in space: Challenges and mitigation techniques,” IEEE Commun. Surveys T uts., vol. 19, no. 1, pp. 57–96, 1st Quart. 2016. [127] M. Sto jano vic and J. Preisig, “Underwater acoustic communi- cation c hannels: Propagation mo dels and statistical character- ization,” IEEE Commun. Mag., vol. 47, no. 1, pp. 84–89, Jan. 2009. [128] S.-S. Raymond, A. Abubakari, and H.-S. Jo, “Co existence of power-con trolled cellular net works with rotating radar,” IEEE J. Sel. Areas Commun., vol. 34, no. 10, pp. 2605–2616, Oct. 2016. [129] D. Chen, S. V akalis, and J. A. Nanzer, “Millimeter-wa ve imaging using a rotating dynamic an tenna array and noise illumination,” IEEE T rans. Antennas Propag., vol. 70, no. 11, pp. 10 965–10 973, Nov. 2022. [130] J.-R. Jiang, C.-M. Lin, and Y.-J. Hsu, “Lo calization with rotatable directional antennas for wireless sensor netw orks,” in Proc. IEEE Int. Conf. Parallel Pro cess. W orkshops (ICPPW), San Diego, CA, Sep. 2010, pp. 542–548. [131] F. Ali, G. Bauer, and M. V ossiek, “A rotating syn thetic aperture radar imaging concept for robot na vigation,” IEEE T rans. Microw. Theory T echn., vol. 62, no. 7, pp. 1545–1553, Jul. 2014.