《MIMO多进多出多天线》课程教学资源（文献资料）6G——Massive MIMO is a Reality—What is Next? Five Promising Research Directions for Antenna Arrays

Massive MIMO is a Reality-What is Next? Five Promising Research Directions for Antenna Arrays Emil Bjornson,Luca Sanguinetti,Henk Wymeersch,Jakob Hoydis,Thomas L.Marzetta 64 fully a reality.Bas Antenna il8eant 6 required and the limitation due【 directio Digital signal iewed processing (b)Beamforming to one point in space ith fyd ream 1:Beamforming from an antenna array can be usedt of the s the r angular or (b bea un.The coming wide- scale deployment of BS the might have no dominant directivity.The radiation pattems in this figure were computed using eight tial pro ing are omnipre antenna uniform linear arrays. tion applic ns,such as low-power ach ing and ning.We ou dominant directivity as shown in Fig.1(b).Both are commonl referred to if ae ape .Si-di is strictly speaking only created in the former case.In addition he aray an t desed to sen the p ions,po ing. er [1]by I.INTRODUCTIO exploration,as While an individual nta has a f n antenna arrays are capa th ing on pat This is traditionally illustrated as the formation of spatial as maximum ratio (MR).zero-forcing (ZF).and minimum beams in one (or a few)distinct angular directions,as shown eror(MMSE)processing were already know in Fig.I(a) Fays are a O C many 0 ng thi that controls be used to focus a signal at an iple antenna communications are still treating MR ZE and arbitrary point in space which.in a rich multi-path propa MMSE as the state-of-the-art methods.With that in mind,one what has the research community been doing he ams so th 30 letails.E ard very on itsuniqu d it is E.Bjon initial concept and a successful commercial solution.Let us E k at the development of multi-use r MIMO. of Pisa.Dip use an taly (lu i.it) stems that at the same time-f esource.In a paper from 1987 of Tech logy.413 s-be-ab Bell 5)Winters described that one can use antenna arrays to uplink signal from different users 61.Suproc how con

1 Massive MIMO is a Reality—What is Next? Five Promising Research Directions for Antenna Arrays Emil Björnson, Luca Sanguinetti, Henk Wymeersch, Jakob Hoydis, Thomas L. Marzetta Abstract—Massive MIMO (multiple-input multiple-output) is no longer a “wild” or “promising” concept for future cellular networks—in 2018 it became a reality. Base stations (BSs) with 64 fully digital transceiver chains were commercially deployed in several countries, the key ingredients of Massive MIMO have made it into the 5G standard, the signal processing methods required to achieve unprecedented spectral efficiency have been developed, and the limitation due to pilot contamination has been resolved. Even the development of fully digital Massive MIMO arrays for mmWave frequencies—once viewed prohibitively com￾plicated and costly—is well underway. In a few years, Massive MIMO with fully digital transceivers will be a mainstream feature at both sub-6 GHz and mmWave frequencies. In this paper, we explain how the first chapter of the Massive MIMO research saga has come to an end, while the story has just begun. The coming wide-scale deployment of BSs with massive antenna arrays opens the door to a brand new world where spatial processing capabilities are omnipresent. In addition to mobile broadband services, the antennas can be used for other communication applications, such as low-power machine-type or ultra-reliable communications, as well as non-communication ap￾plications such as radar, sensing and positioning. We outline five new Massive MIMO related research directions: Extremely large aperture arrays, Holographic Massive MIMO, Six-dimensional positioning, Large-scale MIMO radar, and Intelligent Massive MIMO. Index Terms—Massive MIMO, future directions, communica￾tions, positioning and radar, machine learning. I. INTRODUCTION While an individual antenna has a fixed radiation pattern, antenna arrays are capable of changing their radiation patterns over time and frequency, for both transmission and reception. This is traditionally illustrated as the formation of spatial beams in one (or a few) distinct angular directions, as shown in Fig. 1(a), but antenna arrays are also capable of many other types of spatial filtering. For example, the signal processing that controls the array can be used to focus a signal at an arbitrary point in space which, in a rich multi-path propa￾gation environment, corresponds to emitting a superposition of many angular beams so that the radiated pattern has no E. Björnson is with the Department of Electrical Engineering (ISY), Linköping University, 58183 Linköping, Sweden (emil.bjornson@liu.se). He was supported by ELLIIT and CENIIT. L. Sanguinetti is with the University of Pisa, Dipartimento di Ingegneria dell’Informazione, 56122 Pisa, Italy (luca.sanguinetti@unipi.it). H. Wymeersch is with the Department of Electrical Engineering, Chalmers University of Technology, 412 96 Gothenburg, Sweden (henkw@chalmers.se). J. Hoydis is with Nokia Bell Labs, Paris-Saclay, 91620 Nozay, France (jakob.hoydis@nokia-bell-labs.com). T. L. Marzetta is with the Department of Electrical and Computer Engi￾neering, New York University, Tandon School of Engineering, Brooklyn, NY (tom.marzetta@nyu.edu). (a) Beamforming in one angular direction (b) Beamforming to one point in space Antenna array Digital signal processing Fig. 1: Beamforming from an antenna array can be used to (a) focus the radiated signal in one angular direction or (b) focus the signal at one particular point in space, in which case the radiated signal might have no dominant directivity. The radiation patterns in this figure were computed using eight￾antenna uniform linear arrays. dominant directivity, as shown in Fig. 1(b). Both examples are commonly referred to as beamforming, even if a “beam” is strictly speaking only created in the former case. In addition, the array can be used to sense the propagation environment, for example, to detect anomalies or moving objects. Many different applications for antenna/sensor arrays have been conceived over the years. The 1988 overview paper [1] by Van Veen and Buckley outlined radar, sonar, communications, imaging, geophysical exploration, astrophysical exploration, and biomedical applications that were identified in the 70s and 80s. The signal processing methods that are nowadays known as maximum ratio (MR), zero-forcing (ZF), and minimum￾mean square error (MMSE) processing were already known at that time, but under different names. When writing this article—30 years later—the recent textbooks [2]–[4] on mul￾tiple antenna communications are still treating MR, ZF, and MMSE as the state-of-the-art methods. With that in mind, one might wonder: what has the research community been doing the past 30 years? The devil is in the details. Every application has its unique characteristics and it is hard to bridge the divide between an initial concept and a successful commercial solution. Let us take a closer look at the development of multi-user MIMO, by which we refer to communication systems that use an￾tenna arrays at the BSs to spatially multiplex several users at the same time-frequency resource. In a paper from 1987 [5], Winters described that one can use antenna arrays to discriminate between uplink signals from different users by spatial processing, called receive combining. A few years later [6], Swales et al. described how antenna arrays can be also arXiv:1902.07678v2 [eess.SP] 12 Jun 2019

MI eo meeme mue deplovment can make the total number of multiplexed 2 tepboymea development was largely driven b tion theoretic breakthroughs for multi-user MIMO wer made in the 00s[10]12].The early papers considere perect channe. ation(CS）and its to rovide guidance on how to deal with the imperfec n in (al MIMO can radiate multiple signals (indicated by different colors)that are focused at their respective receivers.as shown 3)Mos in (b). ing in such systems [13].Beamforming based on angle used to spatially multiplex users in the downlink,in which of-arrival (AoA)estimation and quantization codeb case the spatial processing is called transmit precoding. n roughly the“"ighl clas commu but the accuracy was insufficient to control inter-user interference. ly"into the rage area of the transmitter,an antenna can focu 4)The number of antennas was fairly small (which the is insuffi ective at the receiver,without chng ZF or MMSE processing.even with perfect CSI. sin T These factors created a negative attitude against the multi co sequence is that less signal power is ob ed at othe places and one can,therefo ocusother signalstow the user MIMO technology.which has partially remained until ology has changed dramatically during oints in spa ing mu he radiated power.but if M ning will Massive MIMO is a Realiry of th yto the was in ted that Mand are jointly increased 3).That is why is be resolved by equipping the BSs with very large numbers of the preferable opera ting regime multi-user MIMO. ntennas and utilizing time-division duplex (IDD)operation MIM Olpeemieeas th uplink user can have any numbe of antennas and channel conditions Extensive multi-user MIMO field trials were carried out in Each antenna can be built us ive handse the90s[☑and ArrayC Japa the mu lown (SDMA)did not become a commercial successin the 9 nd finds th s deeply rooted in info 0r00s rties of imperfect CSl into acc nt [21.[31. impact on a few key factors The signal processing complexity is manageable f dedi 1)In a time when circuit-switched low-rate voice commu ircuits are designed [16].[17 odebooks which only work well for ngular sparsity and calibrated array structures metrie.Since muli-user MIMO was rathe complicated ,A large numoe of antennas (M 6)leads to an un ind expensive to implem t in the 90s,i was simply fadin ss agal Ss in sma ent to more B multi-user MIMo technology Since classical BSs rely ence even with imperfect CSIif A solid theory for Massive MIMO in block-fading channel on orthogonal time-frequency scheduling.each BS could has been developed in recent years,thanks to the contributions earchers in academia and industry. -user ching the

2 (a) (b) Fig. 2: A classical BS radiates one signal uniformly into its coverage area, as shown in (a). A BS capable of multi-user MIMO can radiate multiple signals (indicated by different colors) that are focused at their respective receivers, as shown in (b). used to spatially multiplex users in the downlink, in which case the spatial processing is called transmit precoding. The difference between classical communication systems and multi-user MIMO can be seen by comparing Fig. 2(a) and Fig. 2(b). Instead of radiating one signal “uniformly” into the coverage area of the transmitter, an antenna array can focus the same signal at its intended receiver. If M antennas are used, an M times stronger signal can (ideally) be achieved at the receiver, without changing the radiated power. The consequence is that less signal power is observed at other places and one can, therefore, focus other signals towards other points in space without causing much interference between transmissions. If K users are spatially multiplexed in the downlink, each user might be allocated only 1/K of the total radiated power, but if M ≥ K, the beamforming will still make the received signal M/K > 1 times stronger than in the classical system. Hence, the overall spectral efficiency [b/s/Hz] of the system grows proportionally to the number of users if M and K are jointly increased [3]. That is why M  K is the preferable operating regime for multi-user MIMO. Note that the word “multiple” in the MIMO acronym refers to the multiple antennas at the BS and the multiple users, while each user can have any number of antennas. Extensive multi-user MIMO field trials were carried out in the 90s [7] and ArrayComm deployed commercial products in Japan [8, Example 10.1]. However, the multi-user MIMO technology (then known as spatial division multiple access (SDMA) [9]) did not become a commercial success in the 90s or 00s. There are many contributing factors and their relative impact is debatable, but we will mention a few key factors: 1) In a time when circuit-switched low-rate voice commu￾nication was the dominant service, it was not the spectral efficiency but the capability of multiplexing a certain number of users per km2 that was the key performance metric. Since multi-user MIMO was rather complicated and expensive to implement in the 90s, it was simply more cost-efficient to deploy more BSs using classical hardware than to invest in the new and rather untested multi-user MIMO technology. Since classical BSs rely on orthogonal time-frequency scheduling, each BS could multiplex a much smaller number of users than a BS supporting multi-user MIMO. Nevertheless, a denser deployment can make the total number of multiplexed users per km2 the same as with a less dense multi-user MIMO deployment. 2) The technological development was largely driven by heuristics and experiments since the first major informa￾tion theoretic breakthroughs for multi-user MIMO were made in the 00s [10]–[12]. The early papers considered perfect channel state information (CSI) and it took many more years for the information theory literature to provide guidance on how to deal with the imperfect CSI that occurs in any practical communication system. 3) Most telecom operators had frequency-division duplex (FDD) licenses at the time and it is hard to acquire downlink CSI that is sufficiently accurate for beamform￾ing in such systems [13]. Beamforming based on angle￾of-arrival (AoA) estimation and quantization codebooks were considered to send beams in roughly the “right” way. This worked rather well for single-user systems, but the accuracy was insufficient to control inter-user interference. 4) The number of antennas was fairly small (M ≈ 8) which is insufficient to achieve the spatial resolution that is necessary for effective interference suppression, using ZF or MMSE processing, even with perfect CSI. These factors created a negative attitude against the multi￾user MIMO technology, which has partially remained until today, even if the technology has changed dramatically during the last decade. A. Massive MIMO is a Reality To address the shortcomings of conventional multi-user MIMO, the Massive MIMO concept was introduced in [14]. It is now well accepted that many of the previous challenges can be resolved by equipping the BSs with very large numbers of antennas and utilizing time-division duplex (TDD) operation and the uplink-downlink channel reciprocity to achieve a com￾munication protocol that supports arbitrary antenna numbers and channel conditions: • Each antenna can be built using inexpensive handset￾grade hardware components [15], which keeps the cost down. • The communication design is deeply rooted in informa￾tion theory and finds the right operating regime by taking the properties of imperfect CSI into account [2], [3]. • The signal processing complexity is manageable if dedi￾cated circuits are designed [16], [17]. • There is no need to rely on AoA estimation or quantiza￾tion codebooks, which only work well for channels with angular sparsity and calibrated array structures. • A large number of antennas (M ≥ 64) leads to an un￾precedented spatial resolution, robustness against small￾scale fading, and the ability to spatially suppress interfer￾ence even with imperfect CSI if M  K. A solid theory for Massive MIMO in block-fading channels has been developed in recent years, thanks to the contributions of many researchers in academia and industry. Some of the key research directions that are now approaching the finish lines

3 8an in Fig.3.There is no massive difference in size since the many low-gain antennas in Massive MIMO must be compared s are 1351 Thi 1410mn two-dimensional configuration also makes the array 988m 15 20MHz up o ch) signaling using QPSK,16-QAM.64-QAM,and 256-OAM The maximum ra 187mm 70 mm for the array 520mm 154mm AAII an Nokia Airscale are two competing product lines.Many tele (a) Fig.3:Comparison of the form factors of(a)the Eri on AIR 6468.64-antenna array:(b)the Kathrein 80010621 antenna nc ding the used the Ma 5 GH (a of their current commercial shipm has either 320r6 since 64 low antennas in (a)are compared with one high- antennas [37].This demonstrates that Massive MIMO is nov gain antenna in(b). 'too complicated and expensive to implement"has finally been disproved. eie18[211 eray effici Since there standardize which signal pro ing me e BS.he maturity of th s underlined networks can change over time.The first Massive MIMO the two as adve on top cover th products are (proba in the aforem inue unde ning and not have enough simultaneously active users to benefit much that the ba ar rstood and from spatial multiplexing.Instead,the telecom operators are n nas been manly obser for impr ng the performance at th [13].often rooted in the negative pastex nces with multi- edg ed in ary pol I) nts in sd firs user MIMO.For example.Massive MIMO has been accu However when the antenna arrays have been deploved the sive to impe d by a softwa 909 the Ma hybrid analog-digital array implementations and complicated nlicated me od m are needed methods are not needed until the number of simultaneo d their rafhic demands urpass the limits of th infeasible to impleme icularly in mmwave bands but orks hav also at sub-6 GHz band t is this correct? improved theirs ectral efficiency in similar ways In 2014.the tes at Lund University sh ed that ully dig "only"a maior enei 331 1m2018 the FCO suppor ssive MIMO p ducts.including I likely be turned into a commercial product in the nex Erics AIR +0 134].This has ante years Hence,even if the first 5G P ducts for mm Wav and dou and is de for 4G ITE so it i ch the matte

3 988 mm 520 mm 187 mm 70 mm 1410 mm 154 mm (a) (b) Fig. 3: Comparison of the form factors of (a) the Ericsson AIR 6468, 64-antenna array; (b) the Kathrein 80010621 antenna panel with 16 dBi directivity. Both arrays support the 2.5 GHz band. The 64 antennas in (a) have varying polarization, in￾dicated by two colors. There is no massive difference in size since 64 low-gain antennas in (a) are compared with one high￾gain antenna in (b). are the spectral efficiency analysis [18]–[21], system design for high energy efficiency [22]–[24], pilot contamination and decontamination [25]–[29], and power optimization [30]–[32]. The maturity of the research on Massive MIMO is underlined by the two recent textbooks on the topic that cover the fundamentals [2] as well as advanced topics [3]. The research in the aforementioned directions can certainly continue under more practical modeling assumptions, but the main point is that the basics are well understood and noncontroversial. Nevertheless, Massive MIMO has been (and still is) met with skepticism and many misconceptions have flourished [13], often rooted in the negative past experiences with multi￾user MIMO. For example, Massive MIMO has been accused of being too complicated and expensive to implement, as if the transceiver hardware technology had not evolved since the 90s. This belief also spurred large investments into “less expensive” hybrid analog-digital array implementations and complicated beam-searching and beam-tracking algorithms that are needed to operate such arrays. The premise for this development is the belief that fully digital transceiver chains are practically infeasible to implement—particularly in mmWave bands but also at sub-6 GHz bands—but is this correct? In 2014, the real-time testbed at Lund University showed that Massive MIMO with 100 fully digital transceiver chains can be implemented using off-the-shelf hardware, requiring “only” a major engineering effort [33]. In 2018, the FCC approved the first line of Massive MIMO products, including the Ericsson AIR 6468 [34]. This product has 64 antennas connected to 64 fully digital transceiver chains in both uplink and downlink, and it is designed for 4G LTE, so it is even a pre-5G product. The AIR 6468 can be used in either the 2.5 GHz or 3.5 GHz band. Compared to a conventional fixed￾beam sector antenna designed for the same bands, the Massive MIMO array is wider but has a smaller height, as illustrated in Fig. 3. There is no massive difference in size since the many low-gain antennas in Massive MIMO must be compared with one high-gain antenna. The 64 antennas are deployed on 4 rows, each containing 8 dual-polarized antennas [35]. This two-dimensional configuration also makes the array compact compared to the large one-dimensional uniform linear arrays that are commonly considered in the academic literature. The AIR 6468 can aggregate up to three carriers (15- 20 MHz each), supports reciprocity-based beamforming, and signaling using QPSK, 16-QAM, 64-QAM, and 256-QAM. The maximum radiated power is 1.875 W per antenna, which corresponds to 120 W in total for the array. The Ericsson AIR 6468 is not unique: Huawei AAU and Nokia Airscale are two competing product lines. Many tele￾com operators started to deploy this type of array in 2018, including the US operator Sprint that even used the Massive MIMO term in its marketing towards the end users [36]. Huawei reported at the Mobile World Congress 2019 that 95% of their current commercial shipments has either 32 or 64 antennas [37]. This demonstrates that Massive MIMO is now a reality for cellular networks operating in conventional sub- 6 GHz bands. Hence, the previous claim of Massive MIMO being “too complicated and expensive to implement” has finally been disproved. Since there is no need to standardize which signal pro￾cessing methods will be used for beamforming and chan￾nel estimation at the BS, the solutions implemented in real networks can change over time. The first Massive MIMO products are (probably) using fairly simple signal processing methods, such as MR for beamforming and least-squares for channel estimation. The reason is that most current cells do not have enough simultaneously active users to benefit much from spatial multiplexing. Instead, the telecom operators are mainly observing a need for improving the performance at the cell edge, so basic beamforming to arbitrary points in space, as illustrated in Fig. 1(b), is the feature that is implemented first. However, when the antenna arrays have been deployed, the spectral efficiency can be improved by a software update that switches to more advanced methods from the Massive MIMO literature [3]. A gradual refinement makes practical sense: less complicated methods are easier to implement, more advanced methods are not needed until the number of simultaneously active users and their traffic demands surpass the limits of the simpler methods, and then the more advanced methods can be sold as feature upgrades. Note that 3G and 4G networks have improved their spectral efficiency in similar ways. At mmWave frequencies, the first experimental verification of fully digital antenna arrays was presented in 2018. NEC has developed a 24-antenna uniform linear array that supports digital beamforming in the 28 GHz band [38]. This prototype will likely be turned into a commercial product in the next few years. Hence, even if the first 5G products for mmWave communications rely on analog or hybrid implementations to quickly reach the market, it is only a matter of time before

digital solutions prevail and become the most cost and ener (a)Compact co-located Massive MIMO arrays efficient implementations,thanks to the fast development in ly go in opposite direction? B.What is Nert? The development of Massive MIMO ommunication tech in the hands of the 0 number of communication. tion alg signal proce and optimiza oms have Modeling simplifications that have beer demia (e.g.block- ding channels with stochasti e fading or determinist with ang e4ncaa0g chance to try out the existing algorithms. arrays,as shown in (a).The e users are in t it is hard to nent is actually nee fomm of a signal beam.To deploy arrays with very man antennas.we can instead create ELAAs where the antennas are distributed over a large area and hidden into other constructior under practical.hardware-related and regulatory constraints ements. At the same time,it is important to initiate more forward- the new signal beams. might If 5G becomes a commercial success,massive digitally a set of users.There practical limits be d in the ind rooftop locations,fo provide spatial multiplexing over wide areas.while new BSs operating in the mmWave bands will be deployed indoors and at the street leve to provide local area The to an the spatial multiplexi atial r waveforms that give constructive interference at particula in the points in space and resolving the hine details of received where large one-dimensional arrays are often considered in a two-dimensional world.In many practical deployment scena ments evolve in the future? s are mainly separable in the nonzonta The remainder of this ar icle will consider five forward- small The sear tions tha aim at us sing antenna arrays existing 64-antenna products have only eight antennas pe exciting ho ontal row massive is tha sector site that ce mul interesting research challenges. the norm in c II.DIRECTION 1:EXTREMELY LARGE APERTURE ARRAYS han a few hundred antennas per site and to obtain a truly The antenna separation in an array is of the order of the ive spatial horizontal domain.we nee velength A and the users are located in the fa -field of the BS array.Ih are two clas mptions in tead of gath hering all the antenas ina single to be reasonablebut be revised gong forward.The a substantially lareer area and made invisible b grows monotonically them into existing construction elements.Fig.4(b)exemplifie antennas 28]. we can expect dreds or thousands of antennas are used te ing me dual-polaned one

4 digital solutions prevail and become the most cost and energy efficient implementations, thanks to the fast development in semiconductor technology. This should come as no surprise— in the time when digitalization is embraced in every part of the society, why would cellular technology suddenly go in the opposite direction? B. What is Next? The development of Massive MIMO communication tech￾nology is now in the hands of the product departments of companies such as Ericsson, Huawei, Nokia, etc. A large number of communication, signal processing, and optimiza￾tion algorithms have been developed over the years and it remains to be seen which ones will work well in practice. Modeling simplifications that have been made in academia (e.g., block-fading channels with stochastic small-scale fading or deterministic channel models with angular sparsity) might prevent a straightforward transfer from theory to practical implementation. Before the product developers have had the chance to try out the existing algorithms, it is hard to tell what further algorithmic development is actually needed. The MIMO research community should certainly support the prod￾uct developers in their efforts to implement existing algorithms under practical, hardware-related and regulatory constraints. At the same time, it is important to initiate more forward￾looking research that considers new applications of antenna arrays that might become the foundation for beyond 5G networks. If 5G becomes a commercial success, massive digitally controllable antenna arrays will be deployed “everywhere”. Conventional sites operating in the sub-6 GHz band will be equipped with arrays of 64 or more antennas (per sector) to provide spatial multiplexing over wide areas, while new BSs operating in the mmWave bands will be deployed indoors and at the street level to provide local area coverage. The network equipment that controls these antennas has access to an unprecedented spatial resolution in terms of emitting waveforms that give constructive interference at particular points in space and resolving the fine details of received waveforms. What else can we use this spatial resolution for, beyond mobile broadband applications, and how will the antenna deployments evolve in the future? The remainder of this article will consider five forward￾looking research directions that aim at using antenna arrays for new non-communication applications and deployment con￾cepts that open up new exciting possibilities but also pose interesting research challenges. II. DIRECTION 1: EXTREMELY LARGE APERTURE ARRAYS The antenna separation in an array is of the order of the wavelength λ and the users are located in the far-field of the BS array. These are two classical assumptions in the array processing and wireless communication literature that used to be reasonable but need to be revised going forward. The spectral efficiency of Massive MIMO grows monotonically with the number of antennas [28]. Thus, we can expect a future where hundreds or thousands of antennas are used to (a) Compact co-located Massive MIMO arrays (b) Extremely large aperture array Fig. 4: The first deployments of Massive MIMO use compact arrays, as shown in (a). The users are in the far-field of the array and, thus, the transmission to a LoS user takes the form of a signal beam. To deploy arrays with very many antennas, we can instead create ELAAs where the antennas are distributed over a large area and hidden into other construction elements, for example, windows as in (b). The user might be in the near-field of the array and then LoS users will not observe signal beams. serve a set of users. There are, however, practical limits to how many antennas can be deployed at conventional towers and rooftop locations, for example, determined by the array dimensions allowed by the site owner, the weight, and the wind load. The rather compact 64-antenna product shown in Fig. 3(a) has already been deployed and we will likely see de￾ployments of somewhat larger arrays at some locations as well. Nevertheless, the spatial multiplexing capability of these two￾dimensional planar arrays in our three-dimensional world is far from what has been demonstrated in the academic literature, where large one-dimensional arrays are often considered in a two-dimensional world. In many practical deployment scenar￾ios, the user channels are mainly separable in the horizontal domain [35] since the variations in elevation angle between different users and scattering objects are relatively small. The existing 64-antenna products have only eight antennas per horizontal row—how massive is that? Since multi-sector sites are the norm in cellular networks, co-location of three or more compact arrays that point in different directions are also likely to happen, as illustrated in Fig. 4(a). However, to deploy more than a few hundred antennas per site and to obtain a truly massive spatial resolution in the horizontal domain, we need new antenna deployment strategies. Instead of gathering all the antennas in a single box, which will be visible and heavy, the antennas can be distributed over a substantially larger area and made invisible by integrating them into existing construction elements. Fig. 4(b) exemplifies a setup where the antennas are deployed next to each window in a tall building. If one dual-polarized antenna is hidden in

each corner of the window,there are 1512 antennas in this Research direction mely large apert山reay MIMO [55 A M056 compact arrays illustrated in Fig.3:the antenna separation is at the order of meters,which is much larger than the wavelength lolographic Massive MIM (be ing in fron de s62 exa mple of an ELAA is when the antennas are distributed nt wa so that each user is essentially ce66 TL: alled Cell-fre eMIM0401-42. TABLE I:The proposed research directions 1 and 2 collect many other research topics as special cases concent has its roots in papers on distributed mimo from th early 00s [43].[44]andc oordinated multipoint from the early Impo tly,the spat an bt th so it is generally beneficial to beamforming at a particular point in space using we ons but are nhase-shifted to add con ution the 48 491 at the target point.Due to the different directivity.the signal ually decays when leaving target poin We use the flaa terminology to iointly des ibe a family gra of arch topics that have previously beer considered sepa- n a s d the tar with radius S 541.The rate owing c that features signal amplification is typically of this size in the BS an nas that are jointly and coherently serving many ear-field while it can be much larger in the far-field,implying distrbuted users. belia list cial cases s is provided in Table A.Vision quence of using elaas is that the radiative near The grand vision of ELAAs is to provide order -of field stretches many kil away magnitude hig oughput in wireless networks com y be in the ary spatial channel ies [391.[501.In the and the distribut ed ante field,the signal that reaches the array from a user s well and inc the appro by a superpos of pla ach in th don (UDN ith an Ao 661.167]that alsc resolve not only the Aoa of a wave but als but each antenna box is then an auton us BS that service has traveled (e.g.the spatial depth)by exploiting the spherical n exclus e set of users.It is known that the throughpu shape of t It is a that rences to visihle to a suhset of the in the this barrier as the number of antennas grows large [281.[29 blocked to the other antennas 51].Hence.channel mo deling [691-1711.at least in theory.The ultimate goal of ELAAs i is substantially o deploy so ma erently ope rating antennas that all the ore p when using EL MAs and ramel hard ilar to to a per- out since there are many antenna vith similar channel gain hannel without any propagation loss [53].[72].In additio exp I the rom ELA epa ate o enhancing mobile broadb services,which c the h grea in 53 which of an unprecedented number of machine-t is known favorable propagation in t ive MIMO devices [3].Anothe quenc rans The mas production of smartphones has tured adva om at dis atly the nid deve ent of Sim

5 each corner of the window, there are 1512 antennas in this example. Suppose the adjacent windows are 3 m apart, then the array spans an area of 24 m × 60 m. This is an Extremely Large Aperture Array (ELAA) [39] compared to the conventional compact arrays illustrated in Fig. 3; the antenna separation is at the order of meters, which is much larger than the wavelength (being in the range from one decimeter down to a few millimeters in the frequency ranges considered in 5G). Another example of an ELAA is when the antennas are distributed over a large geographical area so that each user is essentially surrounded by BS antennas, rather than the conventional case of each BS being surrounded by users. This has recently been called Cell-free Massive MIMO [40]–[42], but the basic concept has its roots in papers on distributed MIMO from the early 00s [43], [44] and coordinated multipoint from the early 10s [45]–[47]. Importantly, the spatial resolution of an array is not determined by the number of antennas but the array’s aperture, so it is generally beneficial to spread out antennas, even if this also gives rise to spatial aliasing phenomena where signals coming from widely different directions cannot be separated [3]. Non-uniform array geometries can provide better spatial resolution than uniform geometries [48], [49]. We use the ELAA terminology to jointly describe a family of research topics that have previously been considered sepa￾rately but all comply the following definition. Definition 1: An ELAA consists of hundreds of distributed BS antennas that are jointly and coherently serving many distributed users. A list of different special cases is provided in Table I. We believe that these special cases can, to a large extent, be jointly analyzed under the ELAA umbrella in the future. A consequence of using ELAAs is that the radiative near- field stretches many kilometers away from the array. Hence, the users will typically be in the near-field of the array, instead of the far-field as is traditionally the case, leading to non￾stationary spatial channel properties [39], [50]. In the far- field, the signal that reaches the array from a user is well approximated by a superposition of plane waves, each being determined by two parameters: a channel gain and an AoA. In contrast, an array with an extremely large aperture can resolve not only the AoA of a wave but also the distance it has traveled (e.g., the spatial depth) by exploiting the spherical shape of the wave and/or the channel gain differences to the antennas. It is also possible that some wave components are only visible to a subset of the antennas in the array and blocked to the other antennas [51]. Hence, channel modeling is substantially harder when using ELAAs and involves many more parameters. While conventional Massive MIMO benefits from channel hardening, where the small-scale fading average out since there are many antennas with similar channel gains, we cannot expect the same from ELAAs since well separated antennas have large gain differences [52]. On the other hand, the great spatial resolution will likely make the channels to different users nearly orthogonal [50], [52], [53], which is known as favorable propagation in the Massive MIMO literature [3]. Another consequence is that the transmission from the array cannot be illustrated as a beam, but rather as strong coherent signal amplification at distinct points in space, Research direction Special cases Extremely large aperture array Cell-free Massive MIMO [40] Coordinated multipoint [47] Very large aperture Massive MIMO [55] Distributed MIMO [56] Radio stripes [57] Network MIMO [58] Holographic Massive MIMO Holographic RF system [59] Holographic beamforming [60] Large intelligent surface [61] Reconfigurable reflectarrays [62] Intelligent walls [63] Software-controlled metasurfaces [64] Intelligent reflecting surface [65] TABLE I: The proposed research directions 1 and 2 collect many other research topics as special cases. as illustrated in Fig. 4(b), while the antennas’ signal compo￾nents add non-coherently at most other places. When aiming the beamforming at a particular point in space using well separated antennas, the signal components arrive from widely different directions but are phase-shifted to add constructively at the target point. Due to the different directivity, the signal amplification gradually decays when leaving the target point, but irrespective of the antenna configuration it is always strong in a sphere around the target with radius λ/8 [54]. The volume that features signal amplification is typically of this size in the near-field while it can be much larger in the far-field, implying that closely located users can be spatially multiplexed with less mutual interference when using an ELAA. A. Vision The grand vision of ELAAs is to provide orders-of￾magnitude higher area throughput in wireless networks com￾pared to what Massive MIMO with compact arrays can practi￾cally deliver. The keys to reaching this goal are the even larger number of antennas and the distributed antenna deployment that reduces the average propagation loss and increases the spatial resolution (particularly, in the horizontal domain). There is a competing concept of ultra-dense networks (UDN) [66], [67] that also relies on distributed antenna deployment, but each antenna box is then an autonomous BS that services its own exclusive set of users. It is known that the throughput of UDNs is fundamentally limited by inter-cell interference [68]. Cooperation between the distributed antennas can break this barrier as the number of antennas grows large [28], [29], [69]–[71], at least in theory. The ultimate goal of ELAAs is to deploy so many coherently operating antennas that all the users have mutually orthogonal channels, leading to a per-user throughput similar to that of an additive white Gaussian noise channel without any propagation loss [53], [72]. In addition to enhancing mobile broadband services, which constitute the majority of the wireless traffic [73], the great spatial resolution of an ELAA can also be exploited for spatial multiplexing of an unprecedented number of machine-type communication devices. The mass production of smartphones has turned advanced antennas and transceiver equipment into a commodity. Simi￾larly, the rapid development of integrated circuits (following

Moore's law)has resulted in tiny p time synchronized 177 computationt yMo procne i117)These ing a non-Iss B.Open Problem Th Since ELAAs provide a particular type of spatial channel the co sin a system with thousands of coop ratine correlation,the spectral and energy effi ciency can be com antennas [17],[54].One potential way to re olve this issue results on Mass MIMO with c s to in 28 sor for coding and decoding of the data pplied least in theory MR nre cessing is ver signals.This proc on nient for dis uted deployments sin there is no in the har h The way h ed to g ZF or MMSE 42 trees:there is an electrical connector at one side of the cable Hence.the first major open problem isto implemen the cable can be bent and shaped as you like,no permission is ining scheme uch a quired to deploy it,a d th s to 4(b) ing.in a di multiple radio strip for exan le on might only be we rthwhile to let a part of the ELAA serve per floor in the building.There is no need for fine-tuning the [39.It is not the computationl physica of the intenn when on an abundanc front- eve heir ba eband s S to stripes is that a larg number of antennas share a serial front ocated at a remote location such as an (edge)cloud 1741.181 hau ction to the baseband pro instead of having develop theoreti proce 741 Me pCe mes,circu nta Thi can greatly reduce the cost of the fronthaul infr When using a compact planar as in Fig.3(a).to serve with on ELAA of the type in Fig.4(b) array 28L.42 erenc of the in plan. Each antenna in an ElAa can he constructed using helpful when creating codebooks with beams that the users ins ead of industry-grade lec from,whi is common for system BSs i no It is like a divide and co ach to Bs der wavefronts might not be plane. Second the ant be arbitrarily deployed the array respons vectors ar W-per-ante ante nas in a pr ers n the r-hel he the er is Iw in both cases.Similarly. the uplink signal important but challengin problems.Measurement campaisn noise ratio improved by collecting and improved channel m leling based on physical pr with ray- are I the t channel gain If we to define standardized evaluation scenarios with deterministi an ELAA and a UDN with the same antenna locations. the ELAA achieves stronger receive gnals and less interferenc in compac the af divid has.for example bee en der strated nume rically for cell-free machine-type communications.it is hard to predict its impact networks in [4042 To achieve these gain ennas on ultra-reliable low atency communication services.On the in the hand, n e and ithm hased the air sio mall-scale fading including the sienal hlockases 、that car is described in 57]. indicating that this might be as occur at mmWave frequen s.On the other hand,if we chip scale atomi cks that ne c deplo t in Fig.4a)witth (RF component in Fig.4(b),the latter might be more

6 Moore’s law) has resulted in tiny processors that are incredibly computationally capable, essentially making the computational complexity of MIMO processing a non-issue [16], [17]. These facts can be exploited to achieve a cost-efficient deployment of ELAAs. The challenge lies in the interconnect of all the components in a system with thousands of cooperating antennas [17], [54]. One potential way to resolve this issue is to integrate the antennas and frontends into cables that can be attached to the facade of buildings and then connected to a baseband processor for coding and decoding of the data signals. This processor can either be physically close to the array or located in the basement or cloud. This type of cable is called radio stripes in [57]. The concept is in many ways analog to the string lights that are used to light up (Christmas) trees; there is an electrical connector at one side of the cable, the cable can be bent and shaped as you like, no permission is required to deploy it, and the system continues to operate even if one component breaks down. The deployment in Fig. 4(b) can be achieved using multiple radio stripes, for example, one per floor in the building. There is no need for fine-tuning the physical location of the antennas when one has an abundance of them; in fact, an irregular deployment might even enhance the spatial separability of the users. Another benefit of radio stripes is that a large number of antennas share a serial front￾haul connection to the baseband processor, instead of having a separated connection per antenna, as in the pCell technology [74] and initially considered for Cell-free Massive MIMO [40]. This can greatly reduce the cost of the fronthaul infrastructure. ELAAs will work efficiently even in a cellular deployment, with one ELAA of the type in Fig. 4(b) per cell, if MMSE processing methods are used to cancel inter-cell interference [28], [42]. Each antenna in an ELAA can be constructed using smartphone-grade hardware, instead of the industry-grade hardware used in contemporary BSs, and there are prospects of using even lower-grade hardware to cut cost [75], [76]. It is like a divide-and-conquer approach to BS deployment; if we compare 1 W transmission from a single antenna with 1 M W-per-antenna transmission from M antennas in an array, the latter can lead to M times stronger received signal at the user due to the signal focusing, even if the total output power is 1 W in both cases. Similarly, the uplink signal￾to-noise ratio (SNR) is improved by collecting more signal energy with an ELAA. This is also where ELAAs differ fundamentally from UDNs, where each user is served only by the antenna that gives the largest channel gain. If we compare an ELAA and a UDN with the same antenna locations, the ELAA achieves stronger received signals and less interference since it coherently combines signals from many antennas to achieve the aforementioned divide-and-conquer gains. This has, for example, been demonstrated numerically for cell-free networks in [40]–[42]. To achieve these gains, the antennas in the ELAA need to be phase-synchronized, which is more complicated to achieve in an ELAA than in a compact array. A synchronization algorithm based on over-the-air signaling is described in [57], indicating that this might be a solvable problem. There are also chip-scale atomic clocks that can be used to keep distributed radio-frequency (RF) components time synchronized [77]. B. Open Problems Since ELAAs provide a particular type of spatial channel correlation, the spectral and energy efficiency can be com￾puted using known results on Massive MIMO with correlated fading [3], [20], [28], [78]. The standard channel estimation, transmit precoding, and receive combining schemes can be readily applied—at least in theory. MR processing is very convenient for distributed deployments since there is no need to share channel knowledge between the antennas in the array [40] but this benefit comes at a huge performance loss compared to using ZF or MMSE processing [28], [42], [79]. Hence, the first major open problem is to implement interference-rejecting precoding/combining schemes, such as ZF and MMSE processing, in a distributed or hierarchical way. When the channel gain variations are large over the array, it might only be worthwhile to let a part of the ELAA serve each user [39], [80]. It is not the computational complexity that is the issue but the front-haul capacity requirements that would be extreme if thousands of antennas need to send their baseband samples to a common processing unit, possibly located at a remote location such as an (edge) cloud [74], [81]. It is important to develop theoretical distributed processing architectures, optimized resource allocation schemes, circuit implementations, and prototypes. Channel modeling is another open problem for ELAAs. When using a compact planar array, as in Fig. 3(a), to serve LoS users located in the far-field, the array response vector can be computed as a function of the azimuth and elevation angles of the incoming plane wave [3, Section 7]. This is helpful when creating codebooks with beams that the users can select from, which is common for systems operating in FDD mode. It is challenging, if not impossible, to do the same for ELAAs. First, the users are in the near-field thus the wavefronts might not be plane. Second, the antennas can be arbitrarily deployed so the array response vectors are unknown, a priori, even for users located in the far-field. Hence, the modeling of channels, as well as the acquisition of the channel features in a particular deployment scenario, are important but challenging problems. Measurement campaigns and improved channel modeling based on physical properties and ray-tracing are needed. If the near-field behaviors turn out to be too complicated to model statistically, it will be necessary to define standardized evaluation scenarios with deterministic channels. Prototyping is also an important challenge and will naturally be more complicated and costly than in compact MIMO systems. While an ELAA is very suitable for broadband and machine-type communications, it is hard to predict its impact on ultra-reliable low-latency communication services. On the one hand, a large number of distributed antennas and an extreme spatial resolution can bring macro diversity against small-scale fading, including the signal blockages that can occur at mmWave frequencies. On the other hand, if we compare the conventional deployment in Fig. 4(a) with the ELAA in Fig. 4(b), the latter might be more susceptible to

large-scale shadowing.Clearly.the reason for deploying BSs in(2)from the latter two.Hence.a()is reconstructed a elevated arge the .In et h .opuica ent but case performance.unless the deployment locations,antenna ase,we can instead illuminate it by density.and operation are well optimized for ultra-reliability. the conjugate e III.DIRECTION 2:HOLOGRAPHIC MASSIVE MIMO minating wave is a pilot waveform emitted by a user device The capacity of Massive MIMO grows monotonically with antennas 3 so wave might be generate suaCe the num t n in a practically viable way?The previous section described sing Mr precoding(also known as conjugate beamforming). e option:physically very large arrays of classicante that are sm Two approaches are currently being taken to approximately Iand well sep realize a continuous microwave aperture: opti 1)The first nach use in the form of a spatially continuous transmitting/receiving aperture. This requires a radically new way of designing using a dense array of conventional fraction-of-spaced mght not ante the g e result will be aphic RF System[59例 are gen mixin differenee the desired RFcarrier frequ ,84 2)The second approach mixes an RF reference signal with num of antennas is the asymptotic limit of Massive MIMO. How can we model communication when using a contin implementation described]uses a single RFpor the receptio on the of the which is connecte the defining characteristic of optical holography where the received field is recorded and later reconstructed.The Fig 6(a).Using the holographic terminology,the RF recording medium a photog phic emulsion)only re port generates the refer ence wave and the ution he tial din h。 network is the hol ographic display.If multip the phase is lost.The holographic rece position of"eams"can be tra ived to process trated in Fig.5.circumvents this issue by mixing enable spatial multiplexing.This is conceptually similar wn n ference wave.Suppose to the hybrid ar a(.)+e+)reaches the [861 and medium.which will record design instead of the complicated beam-training proce 1b(红，2=la(红，)+ea+P dures commonly considered for hybrid beamforming =a(红，P+1+2(a(红，)e-iar+w）.Q) Another embodiment of rec which are :su The last t d he faces that are not actively emitting RF signals.but consis many discrete meta-atom with electronically steerab e是shin d by a ignal sent from another location.the metasurface can control In pr ei(o+)b(,y)2 =a(，+a'(,)e2ax+8 in the ven fre ncy band.Metasurfaces are also knowr +(la(红，gP+1)ear+ (2) as intelligent reflecting surfaces [65].[87]and there are eral variations on these names [88191].N Note that the that the RF ed in

7 large-scale shadowing. Clearly, the reason for deploying BSs at elevated locations is to avoid that the signals are blocked by large objects. There is a risk that ELAAs will greatly improve the average user-performance, but degrade the worst￾case performance, unless the deployment locations, antenna density, and operation are well optimized for ultra-reliability. III. DIRECTION 2: HOLOGRAPHIC MASSIVE MIMO The capacity of Massive MIMO grows monotonically with the number of antennas [3] so it would be desirable to have nearly infinitely many antennas, but how can that be deployed in a practically viable way? The previous section described one option: physically very large arrays of classic antennas that are small and well separated to enable essentially invisible deployment. Another option is to integrate an uncountably infinite number of antennas into a limited surface area, in the form of a spatially continuous transmitting/receiving aperture. This requires a radically new way of designing and analyzing antenna arrays; in fact, “array” might not be the right terminology anymore. Research in this direction is taking place under the names of Holographic RF System [59], Holographic Beamforming [60], and Large Intelligent Surface [61], [82]. When a spatially continuous aperture is being used to transmit and receive communication signals, we refer to it as Holographic Massive MIMO since having an infinite number of antennas is the asymptotic limit of Massive MIMO. How can we model communication when using a contin￾uous aperture? Interestingly, the reception and transmission of an electromagnetic field over a continuous aperture is the defining characteristic of optical holography [83], where the received field is recorded and later reconstructed. The recording medium (e.g., a photographic emulsion) only re￾sponds to the intensity |a(x, y)| 2 of the received field a(x, y), where x, y are the spatial coordinates on the detector, thus the phase is lost. The holographic recording/reconstruction process, illustrated in Fig. 5, circumvents this issue by mixing the desired wavefront with a known reference wave. Suppose the plane wave e i(αx+βy) is used as reference wave, then the combined wave b(x, y) = a(x, y) + e i(αx+βy) reaches the medium, which will record |b(x, y)| 2 = |a(x, y) + e i(αx+βy) | 2 = |a(x, y)| 2 + 1 + 2<  a(x, y)e −i(αx+βy)  . (1) The last term in (1) depends on the phase of a(x, y) and the known phase of the reference wave, which means that the phase information has been implicitly recorded. To reconstruct a(x, y), the transparent holographic display is illuminated by a replica of the reference wave, as shown in Fig. 5(b), which yields the reconstructed electromagnetic field e i(αx+βy) |b(x, y)| 2 = a(x, y) + a ∗ (x, y)e i2(αx+βy) + ￾ |a(x, y)| 2 + 1
e i(αx+βy) . (2) If the received wave is of finite angular extent, then a judicious choice of the reference wave separates the first component in (2) from the latter two. Hence, a(x, y) is reconstructed and indistinguishable from the original field. In effect, optical holographic recording constitutes a distributed, homodyne receiver. If the display is not transparent but acts as a mirror in the reconstruction phase, we can instead illuminate it by the conjugate e −i(αx+βy) of the reference wave, which leads to emitting the conjugate wavefront a ∗ (x, y) in the opposite direction. Analogously, in wireless communications, the illu￾minating wave is a pilot waveform emitted by a user device and the reflecting object is the propagation environment, and the reference wave might be generated inside the surface. By emitting the conjugate wavefront from the surface, we can effectively transmit back to the user device using MR precoding (also known as conjugate beamforming). Two approaches are currently being taken to approximately realize a continuous microwave aperture: 1) The first approach uses a tightly coupled array of dis￾crete, active antennas [59]. This can be implemented using a dense array of conventional fraction-of-λ-spaced antennas, connected to RF chains, but the result will be costly and bulky. Alternatively, the RF signals are gen￾erated by mixing two optical signals whose frequency difference equals the desired RF carrier frequency [84]. 2) The second approach mixes an RF reference signal with a large number of nearly passive reflecting elements hav￾ing electronically steerable reflection parameters [62], [85], known as a reconfigurable reflectarray. The specific implementation described in [60] uses a single RF port on the backside of the surface, which is connected to an electronically steerable RF distribution network with radiating elements that emit the wavefront; see Fig. 6(a). Using the holographic terminology, the RF port generates the reference wave and the distribution network is the holographic display. If multiple RF ports are connected to the same distribution network, a super￾position of “beams” can be transmitted and received, to enable spatial multiplexing. This is conceptually similar to the hybrid architectures considered in the mmWave literature (see [86] and references therein), but differs in the use of holographic recording for beamforming design instead of the complicated beam-training proce￾dures commonly considered for hybrid beamforming. Another embodiment of reconfigurable reflectarrays is software-controlled metasurfaces, which are contiguous sur￾faces that are not actively emitting RF signals, but consist of many discrete “meta-atoms” with electronically steerable reflection properties (e.g., shift in phase, polarization, and amplitude) [63], [64]. Hence, when illuminated by an RF signal sent from another location, the metasurface can control the reflections and, for example, beamform the RF signal in a preferable way; see Fig. 6(b). In principle, the metasurface can be reconfigured to behave as an arbitrarily shaped mirror in the given frequency band. Metasurfaces are also known as intelligent reflecting surfaces [65], [87] and there are several variations on these names [88]–[91]. Note that the key difference from the second approach, described in the previous paragraph, is that the RF signal is not generated inside or close

minating wave Reflecting object Virtual image Replica o wave Referenc a(r,) Reflected desired wave ing a ho a(z.y) display Recording Reconstructed desired wavefront (a)Holographic recording (b)Holographic reconstruction g.5 Sketch of theoica hologrpy.Hr the dsired refectedave)ismixed ”to record a(z when it reaches the recording medium.(b)Holographi es reconstructed by optical image of the reflecting obect can be observed. ave,2 RF signal generator 目Receiving user Re transmitter Receiving user (Coniguusrc BF sigal generaor a the backside()omiorfae Ri generad another Fig.6:Two examples of Holographic Massive MIMO.(a)The RF signal is generated at the backside of the surface and propagates through a steerable network to radiat another r location and surfacea gna a cam ()Inc Kr signal is sent elements that generate a beam provid There are three main reasons to consider Holographic Mas. Definition 2:In Holos phic Massive MIMO.an antenna active way to surface with(approximately)continuous aperture is used for conventional discrete antennas.each with dedicated hardware (this is definitely true at optical wavelengths).The numbe RF ports in MO can be equa Our focus here is the motivation for,and the potential electromagnetic waves with arbitrary spatial frequency com ponents, without undesired side-lobes words,the research problems same surface can be reconfigured to work

8 Illuminating wave Reflecting object Reference wave Reflected desired wave Recording medium a(x, y) x y (a) Holographic recording Virtual image Replica of reference wave Reconstructed desired wavefront a(x, y) Same medium acting as holographic display x y (b) Holographic reconstruction Fig. 5: Sketch of the two steps in optical holography. (a) Holographic recording: the desired reflected wave a(x, y) is mixed with a reference wave e i(αx+βy) to record |a(x, y) + e i(αx+βy) | 2 when it reaches the recording medium. (b) Holographic reconstruction: the desired wave is reconstructed by illuminating the medium by a replica of the reference wave, turning the medium into a holographic display. Since the reconstructed wave is perceived identical to the original reflected wave, a virtual optical image of the reflecting object can be observed. RF signal generator Receiving user Surface with radiating elements (a) Contiguous surface with an RF signal generator at the backside. Receiving user RF transmitter Surface with reflecting elements (b) Contiguous surface with an RF signal generated at another location. Fig. 6: Two examples of Holographic Massive MIMO. (a) The RF signal is generated at the backside of the surface and propagates through a steerable distribution network to radiating elements that generate a beam. (b) The RF signal is sent from another location and the metasurface reflects it using steerable elements that generate a beam. to the surface, but relatively far away. We can nevertheless provide the following joint definition. Definition 2: In Holographic Massive MIMO, an antenna surface with (approximately) continuous aperture is used for communication. The surface either actively generates beam￾formed RF signals or control its reflections of RF signals generated at other locations. Our focus here is the motivation for, and the potential benefits of deploying a continuous aperture, and the associated research problems. A. Vision There are three main reasons to consider Holographic Mas￾sive MIMO. First, a continuous aperture system may become a more attractive way to achieve high spatial resolution in wireless communications than having a very large number of conventional discrete antennas, each with dedicated hardware (this is definitely true at optical wavelengths). The number of RF ports in Holographic Massive MIMO can be equal to the number of signals to spatially multiplex. Second, the continuous aperture enables the creation and detection of electromagnetic waves with arbitrary spatial frequency com￾ponents, without undesired side-lobes. In other words, the same surface can be reconfigured to work flawlessly on any

carrier frequency since the classic spatial aliasing effects 3]to the radius of the brain.any MIMO approach would,of antennas essity.entail closely spaced antenna evar has been large-scale discrete e system:generally integrals are more ower transfe convenient than discrete summations achieved ove The vision is that Holographic Mass MIMO can be The hen don by eva and ever havine a high o.force the evane ent wave ntrary to their emit and receive any electromagnetic wav while remaining nature,to transport real power,with high efficiency aesth ically appealing.The extreme spatial resolution nable Super-directivity is nother phenomenon tha is seldon ibly lo By placing Mass MIMO sed to in th achieve array gains far in ex of what conventional arrays behavior of Massive MIMO communication systems.It is and MR pro sing achieve [94].[95]:"For example,conside clearly simpler to analyze a systen continuun transmitting with an aperture 10 wa cated an stea te nu ted in th.Witha gth diar spa ugh The for Hol Ma MIMO the earth it is possible to receive virtually all of the powe radiated by the horn antenna antenna a cre a on t on B Onen Problems al challeng e ted by Holo more antennas and red aheswboaecoan heory [53 796 m. 82]fo perpendicular to the array,have an imaginary-valued wave o scattering propagation would require that the conventiona evanescent waves decay an a they carry only reactiv i.id.Rayleigh model for small-scale fading is discarded ntional far-field wireless replaced by a non and replacing a conventional discrete array with a holographic rch or channel modeling.as well as channel estimation that u array of th physica would add no additional s inh tion structure,is highly needed.The ical to example extreme near-field communication.wireless power ccura utations but these details should be of little con transfer via resonant eva ent wave coupling. arrays The super he communication theorist:the mere existence e of the plan ave rep and the pro ldes th ield nificantly to the electro stic field.An example of extreme nication the Nevertheless.a general framework for pproximate perfo ance analysis remains to be created,and near-field communication is that of an implantable neura e is a risk that simplified models lea 2 48 Mh y digitiz and d to sine 323.8GHZ FSK 60m al rec he done.Fo arrays that generate the RF signals interally.it is importan to have multiple RF ports cor nected to the same surface.so nat users can be spatially muluiplexe the t e hur the ems in on -day munic hes and this leads to mutual coupline 19 which has MIMO te cale fadin If the also e sp y onary, 1[72 pe ling for a pr

9 carrier frequency since the classic spatial aliasing effects [3] that occur when having sub-critically spaced antennas are alleviated by the continuous aperture. Third, a continuous￾space system should be more tractable to analyze than a large-scale discrete-space system; generally, integrals are more convenient than discrete summations. The vision is that Holographic Massive MIMO can be integrated into any surface, including walls, windows, and even fabrics. We can eventually be surrounded by surfaces that can emit and receive any electromagnetic waves, while remaining aesthetically appealing. The extreme spatial resolution enables incredibly low transmit powers and unprecedented spatial mul￾tiplexing capabilities [53]. As a theoretical tool, Holographic Massive MIMO can be also used to investigate the true limit behavior of Massive MIMO communication systems.1 It is clearly simpler to analyze a system comprising a continuum of antennas instead of a countably infinite number of randomly located antennas. The potential use cases for Holographic Massive MIMO go far beyond conventional communications. A classic planar antenna array, utilizing λ/2-spacing on a Cartesian grid, cre￾ates on transmission and detects on reception a discrete set of ordinary (propagating) plane waves, whose three wavenumber components each have a magnitude less than 2π/λ. Keeping the same physical array dimensions, but simultaneously adding more antennas and reducing the spacing brings additional, evanescent, plane-waves into play, that, parallel to the ar￾ray have wave-number magnitude greater than 2π/λ, and perpendicular to the array, have an imaginary-valued wave￾number. Perpendicular to the array, evanescent waves decay exponentially fast in space, and they carry only reactive (imaginary) power. For this reason, evanescent waves do not participate in conventional far-field wireless communications, and replacing a conventional discrete array with a holographic array of the same physical size would add no additional degrees of freedom. Notwithstanding, evanescent waves are critical to some unconventional communication schemes; for example, extreme near-field communication, wireless power transfer via resonant evanescent wave coupling, and super￾directive antenna arrays. These examples are discussed in further detail below. In the extreme near-field, evanescent waves contribute sig￾nificantly to the electromagnetic field. An example of extreme near-field communication is that of an implantable neural transmitter [92], which continuously digitizes and transmits 100 neural channels (48 Mb/s) of data through the scalp to an external receiver a few meters away, using 3.2/3.8 GHz FSK modulation. It is interesting to speculate on the possibility of increasing the number of neural channels by a factor of ten, one hundred, or one thousand—present-day communication theory seems inadequate. Since one wavelength is equivalent 1The existing asymptotic analysis for Massive MIMO with uncorrelated [14] and correlated fading [28] use simplified channel models where the receiver will asymptotically capture more energy than was transmitted [72], which is clearly impossible. Although these results manifest the performance scaling for a practical number of antennas, they cannot be used to study the asymptotic performance limit. Hence, further work is required and [53] provides one step in this direction. to the radius of the brain, any MIMO approach would, of necessity, entail closely spaced antennas. Resonant evanescent wave coupling has been used for wireless power transfer between a pair of coils, at the 10 MHz carrier frequency (30 m wavelength): 60 W of power transfer was achieved over a range of two meters with 40 % efficiency [93]. The near-field is then dominated by evanescent waves. The actions of the receiving coil, tuned to resonance, and having a high Q, force the evanescent waves, contrary to their nature, to transport real power, with high efficiency. Super-directivity is another phenomenon that is seldom considered in communications. By placing antennas close together to realize super-directivity, in principle, one can achieve array gains far in excess of what conventional arrays and MR processing achieve [94], [95]: “For example, consider a transmitting horn antenna, with an aperture about 10 wave￾lengths on a side, located in outer space roughly aimed at the earth. With a one wavelength diameter super-gain antenna on the earth it is possible to receive virtually all of the power radiated by the horn antenna.” B. Open Problems A general challenge presented by Holographic Massive MIMO is that the progress depends on the availability of researchers who are conversant in both communication theory and electromagnetic theory [53], [96], [97]. Holographic Mas￾sive MIMO is investigated in [53], [82] for communication and positioning, respectively, under LoS propagation. An extension to scattering propagation would require that the conventional i.i.d. Rayleigh model for small-scale fading is discarded and replaced by a spatial correlation model, but none of the conventional statistical models for uniform linear arrays with discrete antennas are directly applicable.2 Further research on channel modeling, as well as channel estimation that utilizes its inherent spatial correlation structure, is highly needed. The accurate calculation of the communication performance, in general, requires elaborate numerical electromagnetic com￾putations, but these details should be of little concern to the communication theorist: the mere existence of the plane￾wave representation, and the impedance kernel, provides the essential physics for the formulation of physics-based commu￾nication theory. Nevertheless, a general framework for exact or approximate performance analysis remains to be created, and there is a risk that simplified models lead to incorrect results. The practical implementation of Holographic Massive MIMO has started [60], [62] but more remains to be done. For arrays that generate the RF signals internally, it is important to have multiple RF ports connected to the same surface, so that users can be spatially multiplexed. So far, the transmission from a continuous aperture has been approximated by discrete patches and this leads to mutual coupling [99], which has 2Physics dictates that the small-scale fading satisfies the homogeneous wave equation. If the random field is also required to be spatially stationary, then the power density spectrum for the small-scale fading exists on the surface of an impulsive sphere of radius 2π/λ. Under this model, there is a plane-wave spectral representation for the random field, which yields an efficient method of computing the joint likelihood of the continuum of noisy measurements [98]

10 to be accounted for in the computation of radiated powe new an Physically,the action of feeding current into a patch performs against the created by the patch, itsel and yual to pro equal to a quadratic form of current with computed.The accuracy is thus fundamentally limited by the Hence.ther in the signal generat the mpling rat inte section between hardw are domain.For time-based measurements.comm nly used from 2G until 4G.Fisher information analysis reveals that the rate SNR the andwi distinguished)is limited to the inverse of the bandwidth to find a killer applicatior omething tha be done Hence.high resolution and high a acy can only be obt ng te large The gains over are not c ,101 hased on distances requiring at least four ios m Efficient p otoco and algorithms for channel estimation and rec d properties to 6 thus far only played rginal role With nt the will be inef Ma MIMO Since the phase-shifts are frequency-flat.another challenge is ties for open up.wemightsimu to operate the surfaces over equenc e more antennas and more bandwidth than in previou nology gener be In and challenging field of res surfaces are des and transmit,and itrary wa Ih The ultim al of sham heams without side-lob and to the asymptotic limits of Massive MIMO,in terms of both sional orientation of the user device,represented by rol nd ya cha el capacity The surfaces can als app ns carrier frea cies First of all,the use of arrays at the enables est IV.DIRECTION 3:SIX-DIMENSIONAL POSITIONING (AoD)(n d mlink)plink 102.wh mation analysis shows that the variance of the AoA estimate bically with the number of antennas.w hile th tin mandates to localize emergency calls,every nev generation The quadrati 10312 ng opp s with the n mbe with ho nd 50-20 antennas [3],which dictates how small angular differe be Here,the number o anten In 3G.the larger bandwidths made time-differ nce-of-arrival surements 4G the position ntal d for for elevation.For example,the 64-antenna array depicted ir to-device communication all improved the positioning per Fig.3(a)has eight dual-polarized antennas in the horizontal formance in successive LIE leases for 10MHz c Generally ing is that the nas int commur cation signals with exteral inf mation,for example e be global navigation satellite system (GNSS)or sional array is u

10 to be accounted for in the computation of radiated power. Physically, the action of feeding current into a patch performs work against the field created by the patch, itself, and the field created by the rest of the patches. As a consequence, the power is not equal to a simple integral of magnitude-current￾squared but rather is equal to a quadratic form of current with respect to a positive-definite impedance kernel. Hence, there are non-trivial factors to consider in the signal generation and beamforming design. Many important open problems in the intersection between hardware properties and electromagnetic waves can be found in this area. For surfaces that reflect RF signals generated elsewhere, many implementation concepts with discrete reflection ele￾ments have been developed [64], [85], [100]. Several potential use cases have been identified [64], [65], [90] but it remains to find a killer application—something that cannot be done with existing technology. For example, it has been shown that metasurfaces can provide range extension [87], [88], but the gains over conventional relaying are not convincing [101]. Efficient protocols and algorithms for channel estimation and fast reconfiguration of the reflection properties need to be developed; the narrower the reflected beams are, the more important the estimation and configuration accuracy will be. Since the phase-shifts are frequency-flat, another challenge is to operate the surfaces over frequency-selective channels and to also transmit efficiently to users in the near field. To summarize, Holographic Massive MIMO is an exciting and challenging field of research, where electronically active surfaces are designed and utilized to receive, transmit, and reflect arbitrary waveforms. The continuous aperture can be employed in conventional communication applications to form sharp beams without side-lobes and potentially operate close to the asymptotic limits of Massive MIMO, in terms of both channel capacity and energy efficiency. The surfaces can also enable unconventional applications of wireless, such as highly efficient wireless power transfer. IV. DIRECTION 3: SIX-DIMENSIONAL POSITIONING An important application of any cellular communication system is the ability to spatially locate its users [102]. While the positioning requirements were traditionally determined by mandates to localize emergency calls, every new generation of cellular networks has provided new positioning opportu￾nities [103]: 2G was limited to cell-ID and coarse timing measurements, with horizontal accuracies around 50-200 m. In 3G, the larger bandwidths made time-difference-of-arrival measurements more accurate, while in 4G, the positioning performance was boosted through dedicated reference signals. Techniques such as carrier aggregation, MIMO, and device￾to-device communication all improved the positioning per￾formance in successive LTE releases, with accuracies now being in the 10 m range for 10 MHz carriers [104]. Generally, positioning is an application of statistical signal processing where a position is computed through the fusion of wireless communication signals with external information, for example, obtained by the global navigation satellite system (GNSS) or barometers [105]. To break the 10 m accuracy barrier and support new ap￾plications, disruptive technologies are needed [106]. Let us first recap the fundamentals of positioning. A user’s position is generally obtained through a two-stage process [107]: mea￾surements are first collected and then a position estimate is computed. The accuracy is thus fundamentally limited by the quality of the underlying measurements. The bandwidth plays an important role since it determines the sampling rate and the ability to resolve multipath components in the frequency domain. For time-based measurements, commonly used from 2G until 4G, Fisher information analysis reveals that the measurement variance is inversely proportional to the SNR and the square of the signal bandwidth, while the resolution (the difference in arrival time between two paths that can still be distinguished) is limited to the inverse of the bandwidth [108]. Hence, high resolution and high accuracy can only be obtained when a large bandwidth is available. The accuracy is also improved with more BSs, with three-dimensional positioning based on distances requiring at least four LoS measurements (due to lack of synchronization between the user and the BSs). MIMO has thus far only played a marginal role in wireless positioning since small arrays provide little benefit. With the realization of Massive MIMO, new dimensions and opportuni￾ties for positioning open up. In 5G, we might simultaneously have more antennas and more bandwidth than in previous technology generations. This combination will be instrumental when superior positioning precision is needed. A. Vision The ultimate goal of positioning is to precisely estimate not only the three-dimensional spatial location, but also the three￾dimensional orientation of the user device, represented by roll, pitch, and yaw. We refer to this as Six-dimensional positioning. Our vision is that this can be achieved using antenna arrays with many antennas, particularly when combined with high carrier frequencies. First of all, the use of arrays at the BSs enables estimation of new physical quantities, namely the AoA (in uplink) and angle-of-departure (AoD) (in downlink) [109]. A Fisher infor￾mation analysis shows that the variance of the AoA estimates decreases cubically with the number of antennas, while the variance of the AoD estimates decreases quadratically with the number of antennas [110]. The main reason is that the beamwidth of the array decreases with the number of antennas [3], which dictates how small angular differences can be resolved.3 Here, the number of antennas should be interpreted along the direction we are interested in; that is, in the horizontal direction for azimuth and the vertical direction for elevation.4 For example, the 64-antenna array depicted in Fig. 3(a) has eight dual-polarized antennas in the horizontal 3To be more precise, the beamwidth of the main-lobe is determined by the length of the array relative to the wavelength. Hence, adding more antennas will decrease the beamwidth (i.e., increase the spatial resolution) if the antenna spacing is constant so that the array size is growing. If we are instead adding more antennas into a fixed array size, it is instead the shape of the side-lobes that changes; see [3, Sec. 7.4] for details. 4The beamwidth of the main lobe can be very different in the azimuth and elevation domains, if a one-dimensional or strongly rectangular two￾dimensional array is used