Enabling Joint Communication and Radar Sensing in Mobile Networks -- A Survey
J. Andrew Zhang, Md Lushanur Rahman, Kai Wu, Xiaojing Huang, Y. Jay Guo, Shanzhi Chen, Jinhong Yuan
11 Enabling Joint Communication and Radio Sensingin Mobile Networks - A Survey
Md Lushanur Rahman, J. Andrew Zhang,
Senior Member, IEEE , Kai Wu, Xiaojing Huang,
Senior Member, IEEE ,Y. Jay Guo,
Fellow, IEEE , Shanzhi Chen
Fellow, IEEE , and Jinhong Yuan,
Fellow, IEEE
Abstract —Mobile network is evolving from a communication-only network towards the one with joint communication andradio/radar sensing (JCAS) capabilities, that we call perceptivemobile network (PMN). Radio sensing here refers to informationretrieval from received mobile signals for objects of interestin the environment surrounding the radio transceivers. In thispaper, we provide a comprehensive survey for systems andtechnologies that enable JCAS in PMN, with a focus on worksin the last ten years. Starting with reviewing the work oncoexisting communication and radar systems, we highlight theirlimits on addressing the interference problem, and then introducethe JCAS technology. We then set up JCAS in the mobilenetwork context, and envisage its potential applications. Wecontinue to provide a brief review for three types of JCASsystems, with particular attention to their differences on thedesign philosophy. We then introduce a framework of PMN,including the system platform and infrastructure, three typesof sensing operations, and signals usable for sensing, and discussrequired system modifications to enable sensing on currentcommunication-only infrastructure. Within the context of PMN,we review stimulating research problems and potential solutions,organized under eight topics: mutual information, waveformoptimization, antenna array design, clutter suppression, sensingparameter estimation, pattern analysis, networked sensing undercellular topology, and sensing-assisted secure communication.This paper provides a comprehensive picture for the motivation,methodology, challenges, and research opportunities of realizingPMN. The PMN is expected to provide a ubiquitous radio sensingplatform and enable a vast number of novel smart applications.
Index Terms —Joint communication and radio/radar sensing(JCAS), Dual-functional Radar-Communications, RadCom, Mo-bile networks, Sensing parameter estimation, Clutter suppres-sion, Networked sensing, Sensing-assisted secure communication,Waveform optimization.
I. I
NTRODUCTION
Wireless communication and radar sensing (C&S) havebeen advancing in parallel yet with limited intersections fordecades. They share many commonalities in terms of signalprocessing algorithms, devices and, to a certain extent, systemarchitecture. This has recently motivated significant researchinterests in the coexistence, cooperation, and joint design ofthe two systems [1]–[9].
Md Lushanur Rahman, J. Andrew Zhang, Kai Wu, Xiaojing Huangand Y. Jay Guo are with University of Technology Sydney, (UTS),Global Big Data Technologies Centre (GBDTC), Australia. Email:[email protected]; { Andrew.Zhang; Kai.Wu; Xiaojing.Huang;Jay.Guo } @uts.edu.auShanzhi Chen is with China Academy of Telecommunications Technology(CATT), Beijing, China. Email: [email protected] Yuan is with University of New South Wales. Email:[email protected] The coexistence of communication and radar systems hasbeen extensively studied in the past decade, with a focuson developing efficient interference management techniquesso that the two individually deployed systems can operatesmoothly without interfering with each other [4], [5], [10]–[16]. Although radar and communication systems may beco-located and even physically integrated, they transmit twodifferent signals overlapped in time and/or frequency domains.They operate simultaneously by sharing the same resourcescooperatively, with a goal of minimizing interference to eachother. Great efforts have been devoted to mutual interferencecancellation in this case, using, for example, beamformingdesign in [16], cooperative spectrum sharing in [12], op-portunistic primary-secondary spectrum sharing in [13], anddynamic coexistence in [15]. However, effective interferencecancellation typically has stringent requirements on the mo-bility of nodes and information exchange between them. Thespectral efficiency improvement is hence limited in practice.Since the interference in coexisting systems is causedby transmitting two separate signals, it is natural to askwhether we can use a single transmitted signal for bothcommunications and radar sensing. Radar systems typicallyuse specifically designed waveforms such as short pulsesand chirps, which enables high power radiation and simplereceiver processing [17]. However, these waveforms are notnecessary for radar sensing.
Passive radar or passive sensing is a good example of exploring diverse radio signals forsensing [18]–[20]. In principle, the objects to be sensed canbe illuminated by any radio signal of sufficient power, suchas TV signals [21], WiFi signals [22], and mobile (cellu-lar) signals [23]–[25]. This is because the propagation ofradio signals is always affected by environmental dynamicssuch as transceiver movement, surrounding objects movementand profile variation, and even weather changes. Hence theenvironmental information is encoded to the received radiosignals and can be extracted by using passive radar techniques.However, there are two major limitations with passive sensing.Firstly, the clock phases between transmitter and receiver arenot synchronized in passive sensing and there is always anunknown and possibly time-varying timing offset between thetransmitted and received signals. This leads to timing andtherefore ranging ambiguity in the sensing results, and alsocauses difficulties in aggregating multiple measurements forjoint processing. Secondly, the sensing receiver may not knowthe signal structure. As a result, passive sensing lacks thecapability of interference suppression, and it cannot separatemultiuser signals from different transmitters. Of course, the a r X i v : . [ ee ss . SP ] J un Systems Signal Formats and Key Features Advantages DisadvantagesC&S withSeparatedWaveforms
C&S signals are separated in time,frequency, code and/or polariza-tion;C&S hardware and software arepartially shared. Small mutual interference;Almost independent design of C&Swaveforms. Low spectrum efficiency;Low order of integration;Complex transmitter hardware.
CoexistingC&S
C&S use separated signals butshare the same resource. Higher spectrum efficiency Interference is a major issue;Nodes cooperation and complicated signal pro-cessing are typically required.
Passivesensing
Received radio signals are used forsensing at a specifically designedsensing receiver, external to thecommunication system;No joint signal design at transmit-ter. Without requiring any change toexisting infrastructure;Higher spectrum efficiency. Require dedicated sensing receiver;Timing ambiguity;Non-coherent sensing and limited sensing capa-bility when signal structure is complicated andunknown, e.g., incapable of separating multi-user signals from different transmitters;No waveform optimization.
JCAS
A common transmitted signal isjointly design and used for C&S. Highest spectral efficiency;Fully shared transmitter and largelyshared receiver;Joint design and optimization onwaveform, system and network;“Coherent sensing”. Requirement for full-duplex or equivalent capa-bility of a receiver co-locating with the transmit-ter;Sensing ambiguity when transmitter and receiverare separated without clock synchronization.
TABLE I: Comparison of C&S systems with separated waveforms, coexisting C&S, passive sensing, and JCAS.radio signals are not optimized for sensing in any way.The potential of using non-dedicated radio signals for radarsensing is further boosted by machine learning, in particular,deep learning techniques [7], [26], [27]. With these techniques,traditional radar is evolving towards more general radio sens-ing . We prefer the term radio sensing to radar due to itsgenerality and breadth. Radio sensing here can be widelyreferred to retrieving information from received radio signals,other than the communication data modulated to the signal atthe transmitter. It can be achieved through the measurement ofboth sensing parameters related to location and moving speedsuch as time delay, angle-of-arrival (AoA), angle-of-departure(AoD), Doppler frequency and magnitude of multipath signal,and physical feature parameters (such as inherent patternsignals of devices/objects/activities), using radio signals. Thetwo corresponding processing activities are called sensingparameter estimation and pattern recognition in this paper.In this sense, radio sensing refers to more general sensingtechniques and applications using radio signals, correspondingto video sensing using video signals. Radio sensing involvesmore diverse applications such as object, activity and eventrecognition in Internet of Things (IoT), WiFi and 5G networks[6]. In [7], the authors described the ubiquitous use of wirelesstechnologies such as WiFi, Bluetooth, FM radio and mobilecellular networks, as signals of opportunity in the implementa-tion of IoT. These radio signals are transmitted by an existinginfrastructure and are not specifically designed for the sensingpurpose. In [27], the authors surveyed works on WiFi sensingwhere WiFi signals can be used for people and behaviorrecognition in an indoor environment. In [28], it is shown thatother radio signals, such as RFID and ZigBee, can also beused for activity recognition. These publications demonstratethe strong potentials of using low-bandwidth communicationsignals for radio sensing applications.It is seen that, joint communication and (radar/radio) sens-ing (JCAS, aka, dual-functional radar and communications, or RadCom) [1], [3], [6], [8], [9], [29], [30] is emergingas an attractive solution for integrating communication andsensing into one system. The basic concept of JCAS may betraced back to 1970s, and had been primarily investigated fordeveloping multimode or multi-function military radars. Therehas been limited research on JCAS for domestic systems until2010s. In the past few years, JCAS has been studied basedon both simple point-to-point communications such as vehic-ular networks [9], [31]–[34] and complicated mobile/cellularnetworks [10], [11], [35], [36]. The former can find greatapplications in autonomous driving, while the latter may rev-olutionize the current communication-only mobile networks.JCAS aims to jointly design and use a single transmittedsignal for both communication and sensing. This means thata majority of the transmitter modules can be shared by C&S.Most of the receiver hardware can also be shared, but receiverprocessing, particularly the baseband signal processing, istypically different for C&S. Via joint design, JCAS can alsopotentially overcome the two aforementioned limitations inpassive sensing. These properties make JCAS significantly dif-ferent from existing spectrum sharing concepts such as cogni-tive radio, the aforementioned coexisting communication-radarsystems, and “integrated” systems using separated waveforms[2], where communication and sensing signals are separatedin resources such as time, frequency and code, although thetwo functions may physically be combined in one system.In Table I, we briefly compare the signal formats and keyfeatures, advantages and disadvantage of the C&S systemswith separated waveforms, coexisting C&S systems, passivesensing, and JCAS systems.JCAS has the potential to integrate radio sensing intolarge-scale mobile networks, creating what we call
PerceptiveMobile Networks (PMNs) [29], [35], [37]–[39]. Evolvingfrom the current mobile network, the PMN is expected toserve as a ubiquitous radio-sensing network, whilst providinguncompromising mobile communication services. It can be
Event ManagementSmart Mobile DevicesTracking, locatingPersonal Assistance Vehicular Network Environmental SensingSmart HomeSmart City Factory Automation
Drones, UAV
Military Devices
Fig. 1: Use cases of PMNbuilt on top of existing mobile network infrastructure, with-out requiring significant changes on network structure andequipment. It will unleash the maximum capabilities of mobilenetworks, and avoid the prohibitively high infrastructure costsof building separate wide-area radio sensing networks. Witha large coverage, the integrated communication and sensingcapabilities are expected to enable many new applications forwhich current sensing solutions are either impractical or toocostly.
A. Potential Sensing Applications of PMNs
Large-scale sensing is becoming increasingly important forthe growth of our industry and society [29], [37], [38]. Itis a critical enabler for disruptive IoT applications and adiverse range of smart initiatives such as smart cities andsmart transportation [6]. Unfortunately, its adoption is severelyconstrained by the high infrastructure costs due to the limitedcoverage areas of existing sensors. For example, seamlesscamera surveillance over expansive areas will be prohibitivelyexpensive due to the sheer number of cameras and communi-cation links required to connect them. In addition, there aresignificant privacy concerns.PMN is able to provide simultaneous communication andradio sensing services, and it can potentially become a ubiqui-tous solution for radio sensing because of its larger broadbandcoverage and powerful infrastructure. Its joint and harmonisedcommunication and sensing capabilities will increase the pro-ductivity of our society, and facilitate the creation and adoptionof a vast number of new applications that no existing sensorscan efficiently enable. Some earlier work on passive sensingusing mobile signals has demonstrated its potentials. Forexample, [23], [24] and [25] used GSM-based radio signals for traffic monitoring, weather prediction and remote sensingof rainfall, respectively. The perceptive network can be widelydeployed for both communication and sensing applications intransport, communications, energy, precision agriculture, andsecurity, where existing solutions are either infeasible or inef-ficient. It can also provide complementary sensing capabilitiesto existing sensor networks, with its unique features of day-and-night operation and see-through of fog, foliage, and evensolid objects.There have been numerous WiFi sensing demonstrators de-veloped and reported in the literature, for applications coveringsafety, security, health and entertainment [27]. The PMN hasmore advanced infrastructure than WiFi sensing, includinglarger antenna arrays, larger signal bandwidth, more power-ful signal processing, and distributed and cooperative base-stations. In particular, with massive multiple-input multiple-output (MIMO), the PMN equivalently possesses a massivenumber of pixels for sensing. This enables radio devices toresolve numerous objects at a time and achieve sensing resultswith much better resolution.Some of the sensing applications that can be enabled byPMN are illustrated in Fig. 1. More specific examples of novelapplications may include: • Real-time city-wide vehicle classification and tracking,vehicle speed measurement, and on-road parking spacedetection; • Extensive on-street and open space surveillance for secu-rity and safety; • Low-cost automatic street lighting systems; • Fine-granularity environmental sensing including factoryemissions monitoring; • Farm livestock movement and animal migration monitor- ing; • Crowd management for major events and emergencyevacuation; and • Integrated security, safety and health sensing applicationsin households.
B. Contributions and Structure of this Paper
This paper provides a comprehensive survey on the state-of-the-arts research on PMN that realizes JCAS technologyin mobile networks. Different to some existing overviewarticles [1]–[3], [5], [6], [8], we focus on JCAS techniquesthat are tailored to cellular/mobile networks, by providing aclear picture on what the PMN will look like and how itmay be evolved from the current communication-only networkfrom the viewpoints of both infrastructure and technology.In this survey, we consider mobile-specific JCAS challengesand solutions, associated with heterogeneous network archi-tecture and components, sophisticated mobile signal format,and complicated signal propagation environment. We referto complicated mobile signals as those with modulations oforthogonal frequency-division multiple access (OFDMA) andmultiuser-MIMO (or spatial division multiple access, SDMA).We discuss major challenges and required changes to systeminfrastructure for the paradigm shift from communication-only mobile network to PMN with integrated communicationand sensing, and provide a comprehensive review on existingtechnologies and open research problems, to address thesechallenges within the framework of PMN.The rest of this paper is organized as follows. • In Section II, we first discuss the difference betweencommunication and radar waveforms. We then brieflyreview the research on three types of JCAS systems,including realizing communication function in a primaryradar system, realizing radio sensing function in a primarycommunication system, and joint design without beingconstrained to an underlying system. We pay particu-lar attention to how the three types of JCAS systemsovercome the waveform difference to meet the differentrequirements for C&S. Note that, the PMN is an exampleof realizing radio sensing in a primary communicationsystem. • In Section III, we introduce the framework of a PMN, in-cluding system architecture, three types of unified sensingoptions, and signals usable for sensing. • In Section IV, we discuss the required system mod-ifications for realizing sensing on an communication-oriented infrastructure. We review three near-term optionsthat enable JCAS in PMNs without requiring significantnetwork modifications, particularly for time-division du-plexing (TDD) systems. • Section V discusses various major research challenges,as well as research opportunities, in PMNs, includingsensing parameter extraction, clutter suppression, jointdesign and optimization, and networked sensing. • In Section VI, we provide a comprehensive review fortechnologies that have been developed to address thesechallenges and beyond, and remained open research TABLE II: List of abbreviations
Abbreviations Meanings
AoA Angle of arrivalAoD Angle of departureANM Atomic norm minimizationBBU Baseband unitCACC Cross-antenna cross-correlationCRAN Cloud radio access networkCS Compressive sensingCSI Channel state informationCSI-RS Channel state information reference signalsC&S Wireless communication andradar/radio sensingDMRS Demodulation reference signalsFDD Frequency division duplexingGMM Gaussian mixture modelICA Independent component analysisIoT Internet of thingsJCAS Joint communication and radio/radar sensingLFM Linear frequency modulationLFM-CPM LFM-continuous phase modulationMAC Medium accessMI Mutual informationMIMO Multiple-input multiple-outputMMSE Minimum mean-square errorMMV Multi measurement vectormmWave Millimeter waveNR New radioOFDM Orthogonal frequency-division multiplexingOFDMA Orthogonal frequency-division multiple accessPHY PhysicalPAPR Peak-to-average power ratioPCA Principal component analysisPMN Perceptive mobile networkPDSCH Physical downlink shared channelPUSCH Physical uplink shared channelPRB Physical resource-blockRIP Restricted isometry propertyRRUs Remote radio unitsRMA Recursive moving averagingRMSE Root mean square errorSC Single carrierSDMA Spatial division multiple accessSISO Single input single outputSRS Sounding reference signalsSSB Synchronization signal and broadcast blocksSTAP Space-time adaptive processingSVD Singular value decompositionTDD Time-division duplexingUE User equipmentV2V Vehicle to vehicle problems. The research review is organized under eighttopics: mutual information, waveform optimization, an-tenna array design, clutter suppression, sensing parameterestimation, pattern analysis, networked sensing undercellular topology, and sensing-assisted secure communi-cation. We also discuss the technology matureness andresearch difficulty for each topic, and highlight key openresearch problems. • Finally, conclusions are drawn in Section VII.A list of abbreviations used in this paper are provided inTable II.
II. T
HREE T YPES OF
JCAS S
YSTEMS
Based on the design priority and the underlying signalformats, the current JCAS systems may be classified into thefollowing three categories, namely: • Realizing communication function in a primary radarsystem (or integrating communication into radar); • Realizing radio sensing function in a primary communi-cation system (or integrating radar into communication);and • Joint design without being constrained to an underlyingsystem.In the first two categories, the design and research focusare typically on how to realize the other function based onthe signal formats of the primary system, with the principleof not significantly affecting the primary system. The lastcategory considers the design and optimization of the signalwaveform, system and network architecture, without bias toeither communication or sensing, aiming at fulfilling thedesired applications only. PMNs belong to the second class,where communication is already very well realized and themain challenge is how to achieve radar sensing functionalitybased on the existing cellular network infrastructure.Next, we first briefly discuss the major differences betweentraditional communication and radar signals, which are im-portant for understanding the design philosophy of the threecategories of JCAS systems. We then provide a brief reviewon the recent research progress in each of the categories.
A. Major Differences between C&S Signals
Conventional radar systems include pulsed and continuous-wave radars [2], [5], [40]. In pulsed radar systems, shortpulses of large bandwidth are transmitted either individuallyor in a group, followed by a silent period for receivingthe echoes of the pulses. Continuous wave radars transmitwaveforms continuously, typically scanning over a large rangeof frequencies. In either systems, the waveforms are typicallynon-modulated. These waveforms are used in both SISO andMIMO radar systems, with orthogonal waveforms used inMIMO radars [17], [40].In most of radar systems, low peak-to-average power ratio(PAPR) is a desired feature for the transmitting signal, whichenables high efficiency power amplifier and long-range oper-ation. The transmitting waveform is also desired to have anambiguity function with steep and narrow mainlobes, whichis the correlation function of the received echo signals and thelocal template signal [40], [41]. These waveforms are designedto enable low-complexity hardware and signal processing inradar receivers, for estimating key sensing parameters such asdelay, Doppler frequency and angle of arrival. However, theyare not indispensable for estimating these parameters. A pulsedradar receiver typically samples the signal at a high samplingrate twice of the transmitted pulse bandwidths, or at relativelylower sampling rate at the desired resolution of the delay(ranging); while a continuous-wave radar receiver typicallysamples signals at a rate much smaller than the scanningbandwidth, proportional to the desired detection capabilityof the maximal delay. Due to their special signal form and hardware, radar systems generally cannot support very high-rate communications, without significant modifications [8],[41].Comparatively, communication signals are designed to max-imize the information-carrying capabilities. They are typicallymodulated, and modulated signals are typically appendedwith non-modulated intermittent training signals in a packet.To support diverse devices and communication requirements,communication signals can be very complicated. For example,they can be discontinuous and fragmented over time andfrequency domains, have high PAPR, have complicated signalstructures due to advanced modulations applied across time,frequency, and spatial domains.Although being designed without considering the demandfor sensing, communication signals can potentially be used forestimating all the key sensing parameters. However, differentto conventional channel estimation which is already imple-mented in communication receiver, sensing parameter estima-tion requires extraction of the channel composition rather thanchannel coefficients only. Such detailed channel compositionestimation is largely limited by the hardware capability. Thecomplicated communication signals are very different to con-ventional radar and demand new sensing algorithms. There arealso practical limits in communication systems, such as full-duplex operation and asynchronisation between transmittingnode and receiving node, which requires new sensing solutionto be developed. We note that the detailed information on thesignal structure, such as resource allocation for time, frequencyand space, and the transmitted data symbols, can be criticalfor sensing. For example, the knowledge on signal structure isimportant for coherent detection. In comparison, most passiveradar sensing can only perform non-coherent detection withthe unknown signal structure, and hence only limited sensingparameters can be extracted from the received signals withdegraded performance [18], [19].The differences and benefits of JCAS in comparison withindividual radar or communication system are summarized inTable III.
B. Realizing Communication in Primary Radar Systems
Radar systems, particularly military radar, have the extraor-dinary capability of long-range operation, up to hundreds ofkilometers. Therefore, a major advantage of implementingcommunication in radar systems is the possibility of achievingvery long range communications, with much lower latencycompared to satellite communications. However, the achiev-able data rates for such systems are typically limited, due tothe inherent limitation in the radar waveform. In [42], authorsimplemented a combined radar and communication systembased on a software defined radar platform, in which the radarpulses are used for communication. Research work in [5]and [43] shows that, communication network establishmentcan be possible for both static and moving radars used inthe military and aviation domains. Adaptive transmit signalsfrom airborne radar mounted unmanned vehicles can also beused to simultaneously sense a scene and communicate senseddata to a receiver at the ground base station. The objective of
TABLE III: Comparison between Radar, Communication and JCAS
Systems Radar Communication JCAS SystemSignalWaveform
Typically unmodulated single-carrier signals; Pulsed orcontinuous-waveform frequencymodulated; Orthogonal if multiplestreams; low peak-to-averagepower ratio (PAPR) Mix of unmodulated (pilots) andmodulated symbols; Complicatedsignal and resource usage with theuse of OFDMA and multiuser-MIMO techniques; High PAPR. JCAS can use both traditional radar andcommunication signals, with appropriatemodifications.
TransmissionPower
High Low Communications integrated into Radarcan achieve very long link distance.Sensing integrated into a single commu-nication device can only support shortrange, but overall JCAS can cover verylarge areas due to the wide coverage ofcommunication networks.
Bandwidth
Large signal bandwidth. Resolu-tion proportional to bandwidth. Typically much smaller than radar. mmWave signals are very promising forJCAS, due to large signal bandwidthand limited propagation. Sensing appli-cations do not have to rely on largebandwidth, such as known WiFi sensingexamples.
Signal Band
X, S, C and Ku sub-6 GHz and mmWave bands Have an impact on operation distancesand resolution capabilities of JCAS.
TransmissionCapability(Duplex)
Full-duplex (continuous-waveform) or half-duplex (pulsed) Co-located transmitter and re-ceiver typically cannot operate onthe same time or frequency block. Full-duplex is a favourite condition, butnot essential.
Clock Syn-chronization
Transmitter and receiver are clock-locked. Colocated transmitter and receivershare the same timing clock, butnon-colocated nodes typically donot. Clock-level synchronization removesambiguity in sensing parameterestimation, but is not essential forsome sensing applications. such systems is to establish low latency, secure and long-rangecommunications on top of existing radar systems.Realization of communication in radar systems needs to bebased on either pulsed or continuous-waveform radar signals.Hence information embedding is one of the major challenges.For example, in [44], random step frequency signal is used indesigning a JCAS system where the carrier frequency of theradar signal is used for modulating communication informa-tion. In [45], the authors showed that the quasi-orthogonalmulticarrier linear frequency modulation-continuous phasemodulation (LFM-CPM) waveform radiated by a MIMO radarcan be applied for communications with multiple users. Formore information on embedding communication informationto radar signals, the readers can refer to [41] which providesan excellent review on this topic.What is missing here in the literature is the communicationprotocol design and receiver signal processing. Communi-cation protocols, particularly medium access (MAC) layerprotocol and physical layer frame structure, are well designedin communication systems. However, the design of commu-nications protocols which can be fitted into radar signals isnot straightforward. The main challenges lie on the require-ment that communication protocol design shall be seamlesslyintegrated into radar operation. Some early work is reportedin [46], where a frame structure is proposed for JCAS withfrequency-hopping continuous-wave radar signals. Based onthe frame structure, channel estimation techniques are thendeveloped without knowing the frequency hopping sequence at the communication receiver. Nevertheless, a complete receiversignal processing for extracting the information embedded inradar waveform is not well studied yet.
C. Realizing Sensing in Primary Communication Systems
This is the category of JCAS systems that the PMN belongsto, and we will provide a comprehensive survey on it inthe rest of this paper. Here, we briefly review the researchin this category. Considering the topology of communicationnetworks, systems in this category can be classified intotwo sub-categories, namely, those realizing sensing in point-to-point communication systems particularly for applicationsin vehicular networks, and those realizing sensing in largenetworks such as mobile networks..There have been quite a few works on sensing in vehicularnetworks using IEEE 802.11 signals. In [47], the authors im-plemented active radar sensing functions into a communicationsystem with OFDM signals for vehicular applications. Thepresented radar sensing functions involve Fourier transformalgorithms that estimate the velocity of multiple reflectingobjects in IEEE 802.11.p based JCAS system. In [31], auto-motive radar sensing functions are performed using the singlecarrier (SC) physical (PHY) frame of IEEE 802.11ad in anIEEE 802.11ad millimeter wave (mmWave) vehicle to vehicle(V2V) communication system. In [32], OFDM communicationsignals, conforming to IEEE 802.11a/g/p, are used to performradar functions in vehicular networks. More specifically, a brute-force optimization algorithm is developed based onreceived mean-normalized channel energy for radar rangingestimation. The processing of delay and Doppler informationwith IEEE 802.11p OFDM waveform in vehicular networksis shown in [48] by applying the ESPRIT method.There has been rapidly increasing JCAS work reportedfor modern mobile networks. In [49], some early work onusing OFDM signal for sensing was reported. In [50], sparsearray optimization is studied for MIMO JCAS systems. Sparsetransmit array design and transmit beampattern synthesis forJCAS are investigated in [51] where antennas are assignedto different functions. In [52], mutual information for anOFDM JCAS system is studied, and power allocation forsubcarriers is investigated based on maximizing the weightedsum of the mutual information for C&S. In [53], waveformoptimization is studied for minimizing the difference betweenthe generated signal and the desired sensing waveform. In[54], the multiple access performance bound is derived for amultiple antenna JCAS system. In [55], multicarrier waveformis proposed for dual-use radar-communications, for whichinterleaved subcarriers or subsets of subcarriers are assigned tothe radar or the communications tasks. These studies involvesome key signal formats in modern mobile networks, such asMIMO, multiuser MIMO, and OFDM. In [29], [35], [37]–[39],the authors systematically studied how JCAS can be realized inmobile networks by considering their specific signal, systemand network structures, and how radar sensing can be donebased on modern mobile communication signals. Based onreported results in the literature and our own experience andvision on this technology, we provide a comprehensive reviewof existing techniques and open research problems under theframework of PMNs in the following sections.
D. Joint Design Without an Underlying System
Although there is no clear boundary between the thirdcategory of technologies and systems and the previous twocategories, there is more freedom for the former in terms ofsignal and system design. That is, JCAS technologies can bedeveloped without being limited to existing communicationor radar systems. In this sense, they can be designed andoptimized by considering the essential requirements for bothcommunication and sensing, potentially providing a bettertrade-off between the two functions.The mmWave JCAS systems are great examples of fa-cilitating such joint design. On one hand, with their largebandwidth and short wavelength, mmWave signals providegreat potentials for high date-rate communications and high-accuracy sensing. On the other hand, mmWave systems areemerging and are yet to be widely deployed. Millimeter wavebased JCAS can facilitate many new exciting applicationsboth indoor and outdoor. Existing research on mmWave JCAShas demonstrated its feasibility and potentials in indoor andvehicle networks [9], [30], [33], [56]–[60]. The authors in [58]provide an in-depth signal processing aspects of mmWave-based JCAS with an emphasis on waveform design for jointradar and communication system. Future mmWave JCAS forindoor sensing is envisioned in [56]. Hybrid beamforming design for mmWave JCAS systems is investigated in [57].An adaptive mmWave waveform structure is designed in [59].Design and selection of JCAS waveforms for automotive ap-plications are investigated in [60], where comparisons betweenphase-modulated continuous-wave JCAS and OFDMA-basedJCAS waveforms are provided, by analyzing the system modeland enumerating the impact of design parameters. In [9],[33], multibeam technologies are developed to allow C&Sat different directions, using a common transmitted signal.Beamforming vectors are designed and optimized to enablefast beam update and achieve balanced performance betweenC&S.
E. Advantages of JCAS Systems
With harmonised and integrated communication and sensingfunctions, JCAS systems are expected to have the followingadvantages: • Spectral Efficiency:
Spectral efficiency can ideally bedoubled by completely sharing the spectrum available forwireless communication and radar [2], [42], [14], [61]; • Beamforming Efficiency:
Beamforming performancecan be improved through exploiting channel structuresobtained from sensing, for example, quick beam adaptionto channel dynamics and beam direction optimization[62]–[66]; • Reduced Cost/Size:
Compared to two separated systems,the joint system can significantly reduce the cost and sizeof transceivers [2], [3], [50]; • Mutual Benefits to C&S:
C&S can benefit from eachother with the integration. Communication links canprovide better coordination between multiple nodes forsensing; and sensing provides environment-awareness tocommunications, with potentials for improved securityand performance.III. F
RAMEWORK FOR A
PMNIn this section, referring to the work in [29], [35], [37],we present a framework of PMN that integrates radio sensinginto the current communication-only mobile network, usingJCAS technologies. In this framework, we describe the systemarchitecture, introduce three types of unified sensing, anddiscuss communication signals that can be used for sensing.
A. System Platform and Infrastructure
The PMN can evolve from the current mobile network,with modification and enhancement to hardware, systems andalgorithms. In principle, sensing can be realized in either theuser equipment (UE) or base-station (BS). Sensing in UE maymotivate wider end-user applications. Compared to UE, BShas advantages of networked connection, flexible cooperation,large antenna array, powerful computation capability, andknown and fixed locations to enable more reliable sensingresults. Therefore, in the following, we mainly consider BS-side sensing.The evolution to PMN is not limited to a particular cellularstandard. Hence we try to generalize the discussions by con-sidering key components and technologies in modern mobile
Standalone BS
CRAN CentralBBU PoolSensing ProcessingUnit
RRU 1RRU 2RRU 3
Clutter
Uplink Comm. & Uplink SensingClutter Reflection
Fronthaul
Downlink Comm. SignalDownlink Active SensingDownlink Passive Sensing
ClutterRevised as a special UE
Mobile Core
Fig. 2: Illustration of sensing in a PMN with both standalone BS and CRAN topologies. RRU1 is a node supposed to havefull-duplexing capability or equivalent. RRU3 is modified to be a special UE, transmitting uplink signals for uplink sensingin RRU2, with clock synchronization between them. RRU2 can also be modified as a receiver only, to do both uplink anddownlink sensing, as well as communications (receiver only) .networks, such as antenna array, broadband, multi-user MIMOand orthogonal frequency-division multiple access (OFDMA),instead of a specific standard. When necessary, we will referto the 5G new radio (NR) standard.Depending on the network setup, we consider two typesof topologies where JCAS can be implemented, that is, acloud radio access network (CRAN) and a standalone BS.Realization of sensing in a PMN based on these two topologiesis illustrated in Fig. 2. Below we elaborate the system andnetwork setup for the two topologies, and we will then discussthree types of sensing operations based on the topologies insubsection III-B. Requirements for modifying the setup toenable sensing will be discussed in Section IV.
1) CRAN:
A typical CRAN consists of a central unit andmultiple distributed antenna units, which are called remoteradio units (RRUs). The RRUs are typically connected tothe CRAN central via optical fibre. Either quantized radiofrequency signals or baseband signals can be transmittedbetween RRUs and the central unit. As shown in Fig. 2, ina CRAN PMN, the densely distributed RRUs, coordinated bythe central unit, provide communication services to UEs. Theirreceived signals, either from themselves, other RRUs, or fromUEs, are collected and processed by the CRAN central, forboth C&S. The CRAN central unit hosts the original basebandunit (BBU) pool for processing communication functions andthe new sensing processing unit for sensing.A typical communication scenario is as follows: severalRRUs work cooperatively to provide connections to UEs,using multiuser MIMO techniques over the same resourceblocks (same time and frequency slots). In CRAN commu-nication networks, power control is typically applied such thatsignals from one RRU may not reach other RRUs. While it is not necessary, we relax this constraint and assume thatcooperative RRUs are within the signal coverage area of eachother. This assumption is reasonable when dense RRUs aredeployed and used to support surrounding UEs via coordinatedmultipoint techniques. This is not necessary for some typesof sensing as we are going to discuss in next subsection,but it increases the options of sensing. Technically, it is alsofeasible at the cost of increased transmission power even ifonly for supporting sensing, as the downlink signals do notcause mutual communication interference to RRUs.Note that, in this configuration, all RRUs are typicallysynchronized using the timing clock from the GPS signals.This forms an excellent network with distributed nodes forsensing applications.
2) Standalone BS:
The CRAN topology is not necessaryfor realizing sensing in PMNs. A standalone BS can alsoperform sensing using the received signals either from itsown transmitted signals or from UEs. This includes the smallBS that may be deployed within a household, which pushesfor the concepts of edge computing and sensing. Like WiFisensing, such a small BS can be used to support indoor sensingapplications such as fall detection and house surveillance.From now on, our discussions will be referred to the CRANtopology, but most of results are applicable to the standaloneBS one. Hence in the case without causing confusion, we willuse CRAN and BS interchangeably.
B. Three Types of Sensing Operations
There are three types of sensing that can be unified and im-plemented in PMNs, defined as uplink and downlink sensing ,to be consistent with uplink and downlink communications. Inuplink sensing, signals received from UEs are used for sensing, while in downlink sensing, the sensing signals are from BSs.The downlink sensing is further classified as
Downlink ActiveSensing and
Downlink Passive Sensing , for the cases whenan RRU collects the echoes from its own and other RRU-transmitted signals, respectively. The terms active and passiveare used to differentiate the cases of sensing using self-transmitted signals and signals from other nodes. Below, weelaborate each sensing operation.
1) Downlink Active Sensing:
In downlink active sensing,an RRU (or BS) uses the reflected/diffracted signals of itsown transmitted downlink communication signals for sensing.Like a mono-static radar, the sensing receiver is co-locatedwith the transmitter. Downlink active sensing enables an RRUto sense its surrounding environment. Since the transmitterand receiver are on the same platform, they can be readilysynchronized at the clock-level, and the sensing results canbe clearly interpreted by the node without external assistant.However, this setup would require full-duplexing capability orequivalent.
2) Downlink Passive Sensing:
Here, downlink passive sens-ing refers to the case where an RRU uses the receiveddownlink communication signals from other RRUs for sensing.Downlink passive sensing signals will be available to this RRUwhen the transmission power is sufficiently large. In this case,they will always be there together with the downlink activesensing signals, the reflection and refraction of the RRU’s owntransmitted signal. They may arrive at the sensing receiverslightly later than the downlink active sensing signals, due tolonger propagation distances. When all RRUs cooperativelycommunicate with multiple UEs using SDMA, these two typesof signals cannot be readily separated in time or frequency,and therefore sensing algorithms also need to consider down-link active sensing signals if downlink passive sensing is inoperation. In general, downlink passive sensing senses theenvironment between RRUs.
3) Uplink Sensing:
The uplink sensing conducted at theBS utilizes the received uplink communication signals fromUE transmitters. Uplink sensing can be directly implementedwithout requiring change of hardware and network setup.However, it estimates the relative, instead of absolute, timedelay and Doppler frequency since the clock/oscillator is typ-ically not locked between spatially separated UE transmittersand BS receivers. This ambiguity may be resolved with specialtechniques as we will discuss in details in Section VI-E6.Uplink sensing senses UEs and the environment between UEsand RRUs.
4) Comparison:
Downlink sensing can potentially achievemore accurate sensing results than uplink sensing. This isbecause, in the downlink sensing case, RRUs generally havemore advanced transmitters such as more antennas and highertransmission power, and the whole transmitted signals arecentrally known. Additionally, as the sensed results in thedownlink sensing are not directly linked to any UEs, theprivacy issue is largely not a problem. Comparatively, uplinksensing may disclose the information of UE, causing privacyconcerns.Downlink and uplink sensing in PMNs are both feasiblefor practical applications in terms of sensing capabilities. According to the results in [29] and [37], the downlinkand uplink sensing with practical transmission power values(smaller than 25 dBm) can reliably detect objects more than150 and 50 meters away, respectively, in a dense multipathpropagation environment. Additionally, a distance resolutionat a few meters can be achieved for signal bandwidth of 100MHz, an angle resolution of about 10 degrees for a uniformlinear array of antennas, and a resolution of 5 m/s movingspeed within channel coherence period.A comparison of the three types of sensing is provided inTable IV. C. Signals Usable from 5G for Radio Sensing
For 5G NR, we can exploit the following signals for sensing.These communication signals may be further jointly optimizedfor C&S, using methods in, e.g., [10], [11], [67].
1) Signals Used for Channel Estimation:
Deterministic sig-nals specifically designed for channel estimations are availablein many systems. The 5G NR [68] includes the demodulationreference signals (DMRS) for both uplink (Physical uplinkshared channel-PUSCH) and downlink (Physical downlinkshared channel-PDSCH), sounding reference signals (SRS)for uplink, and channel state information reference signals(CSI-RS) for downlink. Most of them are comb-type pilotsignals, circularly shifted across OFDM symbols, and areorthogonal between different users. Especially, DMRS signalsaccompanying the shared channel are always transmitted withdata payload and exhibit user specific features. ThereforeDMRS signals are random and irregular over time, whichrequires sensing algorithms that can deal with such irreg-ularity. Comparatively, signals used for beam managementin connected mode, like SRS and CSI-RS can be eitherperiodic or aperiodic, and hence they are more suitable forsensing algorithms based on conventional spectrum estimationtechniques such as ESPRIT.The number and position of DMRS OFDM symbols areknown to BSs, and they can be adjusted and optimized acrossthe resource grid including slots and subcarriers (resourceblocks). This implies good prospects for both channel es-timation and sensing in different channel conditions. Theallocation of resource grid can be optimized by consideringrequirements from both communications and sensing. With agiven subcarrier spacing, the available radio resources in a sub-frame are treated as a resource grid composed of subcarriersin frequency and OFDM symbols in time. Accordingly, eachresource element in the resource grid occupies one subcarrierin frequency and one OFDM symbol in time. A resource blockconsists of 12 consecutive subcarriers in the frequency domain.A single NR carrier is limited to 3300 active subcarriers asdefined in Sections 7.3. and 7.4 of TS 38.211 in [68]. Thenumber and pattern of the subcarriers that DMRS signalsoccupy have a significant impact on the sensing performance,as we will see in Section VI-E.In [39], some simulation results for both uplink and down-link sensing using DMRS are provided. The signal is generatedaccording to the Gold sequence as defined in [68] of , for both
PDSCH and
PUSCH . The generated physical TABLE IV: Comparison of Three Types of Sensing Operations
Types Signals Action Advantages DisadvantagesDownlink ActiveSensing
Reflects from aRRU/BSs owntransmitted downlinkcommunicationsignal Sense surroundingenvironment of theRRU/BS. All data symbols in the receivedsignals can be used and are cen-trally known. Generally require full duplexoperation and other networkmodifications. Devices can bespecially deployed to resolvethis problem.
DownlinkPassive Sensing
Received downlinkcommunicationsignals from otherRRUs Sense environmentbetween RRUs. RRUs are synchronized. Privacyis less an issue because sensedresults not directly linked to anyUEs.
Uplink Sensing
Uplink communica-tion signals from UEtransmitters Sense UEs andenvironmentbetween UEs andRRU. Require minimum modificationto communication infrastructure.Does not require full-duplexing. Timing and Doppler frequencymeasurement could be relative.Transmitted information signalsare not directly known. Rapidchannel variation when UEs aremoving. resource-block (PRB) is over a 3-D grid comprising a 14-symbol slot for the full subcarriers across the DMRS layersor ports. The interleaved DMRS subcarriers of PDSCH areused in downlink sensing, while groups of non-interleavedDMRS subcarriers of PUSCH are used in uplink sensing.The results demonstrate the feasibility of achieving excellentsensing performance with the use of the DMRS signals.However, a major problem of sensing ambiguity is also noteddue to the interleaved patten of the subcarriers.
2) Non-Channel Estimation Signals:
Several deterministicnon-channel estimation signals such as the synchronizationsignal and broadcast blocks (SSB) can also be used forsensing. Such signals typically have regular patterns witha periodic appearance at an interval of several to tens ofmilliseconds. However, they only occupy a limited numberof subcarriers, which may lead to limited identification ofmultipath delay values.
3) Data Payload Signals:
In addition, we can also exploitthe data payload signals for sensing. In downlink sensing, thedata symbols are known to the sensing receiver and hencecan be directly used. In uplink sensing, symbols need to beused in a decision-directed mode. Since these data symbolsare random and signals in different spatial streams are non-orthogonal, they are not ideal for sensing. If it is used foruplink sensing, the signals need to be demodulated first, whichcould also introduce demodulation error. However, they cansignificantly increase the number of available sensing signals,and hence improve the overall sensing performance at the costof increased complexity. Precoders for these signals can beoptimized by jointly considering the requirements from C&S.IV. R
EQUIRED S YSTEM M ODIFICATIONS
C&S can share a number of processing modules in a MIMO-OFDM transceiver, as illustrated in Fig. 3. The whole transmit-ter and many modules in the receivers that are shown in purpleare shared by C&S. The transmitted signal waveform can beoptimized by jointly consider the requirements for C&S, aswill be detailed in Section VI-B. Note that sensing parameter estimation can be done in both time domain and frequencydomain. The sensing applications may demand either sensingparameter estimation or pattern recognition results, or both.Despite the numerous modules shareable by C&S, somemodifications at hardware and network levels to existingmobile networks are necessary for realizing PMNs. As dis-cussed in Section II-A, communication signals can generallybe directly used for estimating sensing parameters, but thecommunication system platform is not directly ready for sens-ing. On one hand, a communication node does not have thefull-duplex capability at the moment, that is, transmitting andreceiving signals of the same frequency at the same time. Thismakes mono-static radar sensing infeasible without modifyingcurrent communication infrastructure. On the other hand, fortransmitter and receiver in two nodes spatially separated, thereis typically no clock synchronization between them. This cancause ambiguity in ranging estimation, and makes processingsignals across packets difficult. Thus bi-static radar techniquescannot be directly applied in this case. These are fundamentalproblems that need to be solved at the system level, to makesensing in primary communication systems feasible.We now describe the modifications of current hardware andsystems that are required to evolve current communicationonly mobile networks to PMNs. The depicted changes focus onthe fundamental reforms that allow the current mobile networkto do radio sensing simultaneously with communication. Inthis section, we do not consider low-level changes such asjoint waveform optimization [10], [11], [67], joint antennaplacement and sparsity optimization processing and poweroptimization [50], but leave them to Section VI.For uplink sensing, if the sensing ambiguity in time andDoppler frequency can be tolerated, no change to hardwareand system architectures of current mobile systems is required.Otherwise, achieving non-ambiguous sensing in the PMNspotentially requires dedicated (static) UEs that are clock syn-chronized to BSs. Such ambiguity may also be resolved usingsignal processing techniques as will be detailed in SectionVI-E6.For downlink sensing, the leakage and reflected signals Transmitter (Multiuser‐MIMO OFDMA Signal) ChannelRF ProcessingA/D conversionConvert to frequency domainsynchronizationMIMO Equalization Channel EstimationDemodulation and decoding Pattern RecognitionSensing Parameter EstimationSensing Applications
Fig. 3: A block diagram of a transceiver showing the compo-nents that can be shared by C&S. The blocks in purple areshared by C&S; blocks in blue and black are for communica-tions and sensing only, respectively.from the transmitter can cause significant interference to thereceived signals. Resolving this problem would ideally require full-duplex technologies [69]. The full duplex technology,which uses a combination of antenna separation, RF sup-pression and baseband suppression to mitigate the leakagesignal from transmitter to receiver, is a potentially long-termsolution to enable seamless integration of downlink sensingwith communications. However, it is still very challenging toimplement particularly for MIMO system, and the technologyis immature and impractical for real implementations.Referring to Fig. 2, there exist near-term solutions forrealizing JCAS in PMNs, where radio sensing can be realizedin some suboptimal way without full-duplexing, requiring onlya few slight modifications on hardware and system to theexisting network. These solutions are detailed below.
A. Dedicated Transmitter for Uplink Sensing
Conventionally, the phase clock between UEs and BSsis not synchronized; hence, the sensing ambiguity problemis present in uplink sensing. To eliminate the ambiguity,dedicated (static) UEs that are clock-synchronized to BSs canbe used. In terms of the required system modification, uplinksensing by static UE would be the most convenient way forachieving non-ambiguity sensing in the PMNs. This is shownas RRU3 in Fig. 2 for a CRAN, where RRU3 can be modifiedto operate as a UE, transmitting uplink signals.
B. Dedicated Receiver for Downlink Sensing
For downlink sensing without requiring full-duplexing ca-pability, one option is to deploy a BS that only works onthe receiving mode. It can be configured as a receiver eitherfor downlink sensing only or for both communication anddownlink sensing.To implement this near-term downlink sensing, changesto the hardware may be required. This is because the re-ceiver in current BSs is conventionally designed to receive uplink communication signals only, and downlink sensingrequires the receiving of downlink communication signals. Therequired change is insignificant for time-division duplexing(TDD) systems since a TDD transceiver generally uses aswitch to control the connection of antennas to the transmitteror receiver. Thus the change is only the adjustment of thetransmitting and receiving period so that the switch is equiva-lently always connected to the receiver. For frequency divisionduplexing (FDD) systems, the BS receivers may be incapableof working on downlink frequency bands, and modification tothe hardware is required. Therefore, it is more cost-effectiveto implement downlink sensing in TDD than in FDD systems.Alternatively, we can also deploy a dedicated receiving-only node for both downlink and uplink sensing, as wellas communications if desired. This is particularly feasiblefor TDD systems. In TDD systems, downlink and uplinksensing signals can then be (largely) separated in time at thereceiver. Even if this node only has one receiving antenna,we can still use its collected signal to estimate the angle ofdeparting (AoD) values if multiple antennas are applied inthe transmitter with position known to the receiver [70]. Ofcourse, to remove the ambiguity in delay estimation, clocksynchronization is required between the transmitters and thisnode. An example is shown as RRU2 in Fig. 2 for a CRAN,which can perform downlink and uplink sensing using receivedsignals from RRU1 and RRU3, respectively.
C. BS with Spatially Widely Separated Transmitting and Re-ceiving Antennas
One possible solution for downlink sensing is to use well-separated transmitting and receiving antennas. The large sepa-ration will significantly reduce the leakage from transmittedsignals. The receiver baseband also accepts feedback fromthe transmitter baseband, so that a baseband self-interferencecancellation may be further applied. However, this spatiallywell-separated antenna structure requires extra antenna instal-lation space and can increase the overall cost. One option ofminimizing the cost is to use a single antenna for receivingsensing signals.Fig. 4 shows an example of this option in TDD systems.The system has a normal transceiver for communication withfour antennas. A fifth antenna is installed at a position wellseparated from the four antennas, and it is connected to thereceiver via a long cable. Signals from the fifth antenna areused for downlink active sensing. Fig. 4-(a) plots the generalconcept, and Fig. 4-(b) shows a potential implementation inexisting FDD systems. The switches (SPDT1-4) are operatingnormally for a TDD communication system. For the fifthantenna, it is always connected to the fifth receiver. Given thatthe on-board circuit leakage is small and the TDD switchescan be separately controlled, this option can be convenientlyrealized in an existing TDD system that supports 5x5 MIMO.Sensing using a single receiving antenna in this case can berealized by exploring the multiple transmitted spatial streams[70]. Normal Tx/RxSensing receiver only
SPDT1
Tx1
SPDT2SPDT3SPDT4
Rx1 (C)Tx2Tx3Tx4Rx2 (C)Rx3 (C)Rx4 (C)Rx5 (S) (a) (b)
Fig. 4: Simplified TDD transceiver model with a single re-ceiving antenna dedicated to sensing. (a) A general concept;(b) Possible realization on existing hardware platform.V. M
AJOR R ESEARCH C HALLENGES FOR
PMNThere exist a number of challenges in the research anddevelopment of PMN. These challenges are mainly associatedwith realizing sensing on the infrastructure of communicationnetworks, joint design and optimization, exploring the mutualbenefits of communication and sensing via the integration,and sensing in a networked environment. In this section, wewill discuss several major research challenges from the signalprocessing aspect. In next section, we will then review detailedtechnologies and algorithms that have been developed toaddress these challenges, and present remaining open researchproblems and future directions.
A. Sensing Parameter Extraction from Sophisticated MobileSignals
Sophisticated signal structure of mobile networks makessensing parameter estimation in PMNs challenging. A mod-ern mobile network is a complex heterogeneous network,connecting diverse devices that occupy staggered resourcesinterleaved and discontinued over time, frequency and space.Mobile signals are also very complicated because of multiuseraccess, diverse and fragmented resource allocation, and spatialmultiplexing. The communication signals that are also usedfor sensing are randomly modulated using multiuser-MIMOand OFDMA technologies and can be fragmented for eachuser - discontinuous over time, frequency or space. Thisstructure is detailed in our work in [35]. Most existing sensingparameter estimation techniques are not directly applicable tothe PMNs because of such signal structure. For example, activeradar sensing technologies mostly transmit linear FM (LFM)chirp modulated transmitted signals [17]; and most passivebistatic and multistatic radars consider simple single carrierand OFDM signals [18]–[20], [71]. In addition, conventionalspectrum analysis and array signal processing techniques, suchas MUSIC [1] and ESPRIT [49], are not applicable either, asthey require continuous observations that are not constantlyexisting here. As a result, specific sensing techniques needto be developed for estimating sensing parameters from thecomplicated and fragmented signals. Sensing parameters describe the propagation of signals inthe environment and the detailed composition of channels.They typically have continuous but not discrete values. Thusmost existing channel estimation and localization algorithmsare not directly applicable either. Existing channel estimationtechniques developed for modern mobile networks principallyemphasize on estimating composite channel coefficients atquantized discrete grids, and localization mainly focus onthe line-of-sight path and determines the locations of signalemitting objects. However, some recent techniques developedfor channel estimation in millimeter wave systems [72], [73]can potentially be extended and applied to sensing parameterestimation, as will be detailed in Section VI-E.
B. Clutter Suppression
Rich multipath in mobile networks creates another chal-lenge for sensing parameter estimation in PMNs. In a typicalenvironment, BSs receive many multipath signals that areoriginated from permanent or long-period static objects. Thesesignals are useful for communications, but for a fixed BS, theyare generally not of interest for continuous sensing becausethey bear little new information. Such undesirable multipathsignals are known as clutter in the traditional radar literature.Although high-end military radar can simultaneously detectand track hundreds of objects, the capability is built onadvanced hardware such as huge antenna arrays of hundredsand thousands of antenna elements. For a PMN BS with tens ofantennas, the sensing capability largely depends on the sensingalgorithm, which is closely related to the number of unknownparameters. Most existing parameter estimation algorithmsrequire more measurements than unknown parameters and theestimation performance typically degrades with the numberof unknown parameters increasing. Therefore, it is crucial toidentify and remove non-information-bearing clutter signalsfrom the input to a sensing parameter estimator.Clutter suppression techniques for conventional radars arenot directly applicable here because the signals and workingenvironment for the two systems are very different. Typicalradar systems are optimized for sensing a limited number ofobjects in open spaces using narrow beamforming, and clutteris typically from ground, sea, rain, etc. and has notable distinctfeatures [1], [74]–[76]. The well-known algorithms in radarsystems, such as space-time adaptive processing (STAP) [74],[77], independent component analysis (ICA) [75], singularvalue decomposition (SVD) [78] and Doppler focusing [79],are adapted to such scenarios. For communications, narrowbeamforming may occur in emerging millimetre wave systems,but not in more general microwave radio systems due to thelimited number of antennas and the use of multibeam tech-nology to support multiuser MIMO. The signal propagationenvironment in PMNs can also be very complex and differentfrom typical radar working environment. Therefore, existingclutter suppression methods developed for radar systems, e.g.,those in [76], [80], [81], may not directly suit for clutterreduction in PMNs. C. Joint Design and Optimization
One key research problem in JCAS, as well as PMNs, ishow to jointly design and optimize signals and systems forC&S. A number of studies have investigated the impact ofthe waveform and basic signal parameters on the performanceof a joint system, as will be detailed in Section VI-B. Suchwaveform and system parameter optimisation can result inperformance improvement in standalone systems, but it hasless impact compared to those at high levels, i.e., system andnetwork levels.C&S have very different requirements at the system andnetwork levels. For example, in a multiuser MIMO commu-nication system, the transmitted signal is a mix of multi-usersrandom symbols, while ideal MIMO-radar sensing signals areunmodulated and orthogonal [82]. When using an array, radarsensing focuses on optimising the formation and structureof virtual subarrays to increase antenna aperture and thenresolution [83], but communication emphasises beamforminggain and directivity. Such conflicting requirements can makejoint design and optimisation very challenging. More researchis required to exploit the commonalities and suppress theconflicts between the two functions.Another important issue is how C&S can benefit more fromeach other via the integration. This is far from being wellunderstood. Current research has been limited to propagationpath optimization such as the work in [84].
D. Networked Sensing
Integrating sensing into mobile communication networksprovides great opportunities for radio sensing under a cellularstructure. However, research on sensing under a cellular topol-ogy is still very limited. The cellular structure for communica-tion is designed to greatly increase the frequency reuse factorand hence improve spectrum efficiency and communicationcapacity. A cellular sensing network intuitively also increasesfrequency reuse factor, and hence the overall “sensing” capac-ity. On one hand, there is almost no known performance boundfor such cellular sensing networks yet, except for a limitednumber of slightly related works, such as performance analysisfor coexisting radar and cellular communication systems [85]and radar sensing using interfered OFDM signals [49]. Onthe other hand, although research exists on distributed radarand multi-static radar, sensing algorithms that consider andexploit the cellular structure, such as co-cell interference, nodecooperation, and sensing-handover over base-stations, are yetto be developed. The challenge lies in the way to address com-petition and cooperation between different base-stations underthe cellular topology, for both performance characterisationand algorithm development of networked sensing.VI. D
ETAILED T ECHNOLOGIES AND O PEN R ESEARCH P ROBLEMS
As a new platform and network, PMN is still in its very earlystage of research and development. As described in the lastsection, there are a number of challenges to overcome to makeit practical, which also imply great research opportunities.Here we review existing technologies and algorithms that have been developed to address these challenges, organizedunder eight topics. We also discuss open research problemsfor each topic. In Table V, remarks are provided on thetechnology maturity and research difficulty for each topic andhighlights selected key open research problems. The scores formaturity and difficulty are indicative only, as they are basedon our own expertise and experience. Since the major issue inPMN is how to achieve radio sensing without compromisingthe performance of existing communications, we focus onthe issues in realizing radio sensing, leveraging the existingcellular communication infrastructure.
A. Mutual Information
Mutual information (MI) [86] can be used as a tool tomeasure both the radar and communication performance. Tobe specific, for communications the MI between wirelesschannels and the received communication signals can beemployed as the waveform optimization criterion, while forsensing, the conditional MI between sensing channels and thesensing signals can be used [87]–[89]. The usage of MI andcapacity is well known to the communication community. Theusage of MI for radar waveform design can also be tracedback to 1990s [87]. MI has also been used to optimize theperformance of coexisting radar communication systems, e.g.,in [90].Mutual information for JCAS systems has been studiedand reported in a few publications. The work in [91] for-mulates radar mutual information and the communicationchannel capacity for a JCAS system. In [86], radar waveformoptimization is studied for a JCAS system by maximizingmutual information expressions. In [5], the estimation rate,defined as the MI within a unit time, is used for analyzingthe radar performance, together with the capacity metric forcommunications. In [92], authors propose an OFDM waveformoptimized by maximizing a weighted sum of the communica-tion data rates and the conditional mutual information for radardetection in an JCAS system.These available results can be used as good basis for study-ing the mutual information for PMNs, with the considerationof the following signal and system architectures specific toPMNs. • Firstly, the MI formulations for uplink and downlinksensing are different, due to the different knowledge onsignals. In downlink sensing, the symbols are known tothe receiver, and the channels for C&S are correlatedbut are different. For uplink sensing, the symbols areunknown to the receiver and the channels are the same.Hence, from information theory, the optimization targetsand results can be quite different for uplink and downlinksensing. • Secondly, formulations of mutual information need toconsider specific packet and signal structures in cellularnetworks. For example, a packet signal may includetraining sequence and data symbols which will lead todifferent MI formulation and results, as their statisticalproperties are different. In [93], the MI is studied forPMN, considering the frame structure and estimation TABLE V: Technology matureness, research difficulty and selected key open research problems. Higher scores stands for moremature, and more difficult.
Research Topics TechnologyMatureness(1 to 10) ResearchDifficulty (1to 10) Selected Key Open Research Problems
Mutual informa-tion 5 7 • MI formulation specific to PMNs by considering uplink and downlink sensing,and actual signal and packet structure; • Combine MI and other metric such as CRLB of estimators to better characterizeperformance of sensing.Waveformoptimization 6 5 • Waveform optimization for hybrid antenna arrays; • Low-complexity optimization schemes that can be quickly adapt to channelvariation in both C&S; • Multiuser correlation in waveform optimization for uplink sensing.Antenna arraydesign 3 7 • Using virtual array and antenna grouping techniques to achieve balance betweenprocessing gain and resolution in sensing, and diversity and multiplexing incommunications; • Sparse array design and signal processing in PMNs.Clutter suppres-sion 7 5 • Parameter optimization in the recursive moving averaging method; • Low-complexity algorithms for parameter estimation in Gaussian mixed model.Sensing parame-ter estimation 3 8 • Off-grid compressive sensing with discontinuous samples; • Off-grid Tensor signal processing algorithms; • Sensing parameter estimation with clustered multipath channels; • Resolution of sensing ambiguity with asynchronous nodes.Pattern analysis 2 5 • Application-driven problem formulation and pattern analysis; • Environment robust algorithms.Networked sens-ing 1 8 • Fundamental theories and performance bounds for cellular sensing networks; • distributed sensing with node grouping and cooperation.Sensing-assisted securecommunication 2 6 Characterize the Secrecy capacity and develop practical code design methods forinformation encryption using sensing results. errors. The findings from [93] indicate that the optimalsolution for one function (communication or sensing)is generally not optimal for the other, and some trade-off needs to be made, particularly when the require-ments for C&S are very different, for example, whenthe directions of sensing and communications deviatesignificantly. This implies the importance of sensing-motivated user scheduling, i.e., taking user schedulinginto joint optimization of C&S.For sensing, maximizing MI essentially maximizes thechannel information at the sensing receiver, conditional on thesensing signal. But it does not directly reflect how accuratethe sensing parameter estimation can be, as most of theestimators are nonlinear. So it would be closer to practicalsystem performance bounds when other performance metricsare also taken into consideration. Actually, MI has beencombined with other metrics to study the performance of radarsystems. For example, two criteria, namely, maximization ofthe conditional MI and minimization of the minimum mean-square error (MMSE), are studied in [88] to optimize thewaveform design for MIMO radar by exploiting the covariancematrix of the extended target impulse response. In [94], the optimal waveform design for MIMO radar in colored noise isalso investigated by considering two criteria: by maximizingthe MI and by maximizing the relative entropy between twohypotheses that the target exists or does not exist in the echoes.Research for JCAS and PMNs based on these combinedcriteria is still very limited. B. Waveform Optimization
For JCAS, joint waveform optimization is a key researchproblem as the single transmitted signal is used for bothfunctions but the two functions have different requirementsfor the signal waveform. As discussed in Section II-A, tra-ditional radar and communication systems use very differentwaveforms, which are optimized for respective applications.For example, recall that radar uses orthogonal and unmod-ulated pulsed or continuous-waveform frequency modulatedsignals, while in PMNs, typically the signals are random, withmulticarrier modulation and multiuser access. However, thewaveform for one function may be modified to accommodatethe requirements of the other, under joint design and optimiza-tion. The work in [1] is one of the earliest ones that investigatewaveform design for JCAS systems. The waveform design and signal parameters can have a significant impact on the overallperformance of a JCAS system. For example, the numericalanalysis in [32] demonstrates the close linkage between thesensing resolution capabilities and the signal parameters forboth single carrier and multicarrier communication systems.For PMNs, apart from the MI-based waveform optimizationas discussed in Section VI-A, there are two more practicalmethods. One method is optimizing the precoding matrices tomake the statistical properties of the transmitted signals bestsuitable for both C&S. Another method is to add the sensingwaveform to the underlying communication waveform, whileconsidering coherent combination of the two waveforms fordestination nodes. The two methods have respective advan-tages and disadvantages. We elaborate them below.In the first method, the precoding matrix is designed toalter the statistical properties of the transmitted signal. It isparticularly suitable for global optimization of cost functionsjointly formulated for C&S. In [10], waveform optimizationis realized via minimizing the difference between the gen-erated signal and the desired sensing waveform under therestrictions of signal-to-interference-and-noise ratio (SINR)for multiuser MIMO downlink communications. A multi-objective function is further utilized to trade off the similaritybetween the generated waveform and the desired one [11]. In[95], adaptive weighted-optimal and Pareto-optimal waveformdesign approaches are proposed to simultaneously improve theestimation accuracy of range and velocity and the channelcapacity for communication. In [52], the weighting vectorfor subcarriers in OFDM systems is optimized by consideringa multi-objective function involving communication capacityand Cramer-Rao lower bounds for the estimates of sensingparameters. One main disadvantage of this method is that, theprecoding matrix needs to be optimized or redesigned oncethe communication or sensing setup changes.In the second method, basic waveforms can be designedin advance for both C&S, and the two waveforms are thenadded in a way to jointly optimize the performance of C&S.This could be particularly useful for millimetre wave systemswhere directional beamforming is used. One example is avail-able from [9], where a multibeam approach is proposed toflexibly generate communication and sensing subbeams usinganalogue antenna arrays. Optimization of combining the twosubbeams is further investigated in [33]. Although the resultsmay be suboptimal, this method provides great flexibilityand can adapt quickly to changes on the requirements forC&S. Of course, the efficiency of multibeam is related to therequirements of C&S. According to [65], getting the correctsolutions of beam steering and beamwidth adaptation for JCASoperation highly depends on environmental context. Indeed,reflector position, blockage height, motion speed and otherenvironmental context factors could have a significant impacton the efficiency of the multibeam method.For waveform optimization in PMNs, the following specificproblem associated with multiuser access is yet to be consid-ered, particularly for uplink sensing. For downlink sensing,multiuser access and multiuser interference only needs tobe considered for communications, because the transmittedsignals are known to the sensing receiver and the environment to be sensed is common to multiuser signals. Thus waveformoptimization only needs to consider the multiuser aspectfor communication, as studied in [11]. However, for uplink,signals need to be specific to each user for both C&S, becausethe signal propagation environments between different usersand the BS could be different. But these environments couldalso be correlated. Thus waveform optimization in the uplinkis a more challenging task. C. Antenna Array Design
For radio sensing, each antenna with an independent RFchain is like a pixel in the camera. But a radio systemallows more flexible control and processing of both transmittedand received signals. Therefore, there are more designs forantenna arrays in PMNs that we can do apart from theMIMO precoding for waveform optimization as discussed inlast subsection. Below, we exemplify two research topics onantenna array design.
1) Virtual MIMO and Antenna Grouping:
There are manycontradictory requirements for antenna array design betweenC&S. Beamforming and antenna placement are two goodexamples. For beamforming, an array with steerable beam-forming and narrow beamwidth is typically required for sens-ing; however, communications require fixed and accuratelypointed beams to achieve large beamforming gain. For an-tenna placement, increment of antenna aperture is the mainconcern for radar [83], while MIMO communication focuseson beamforming gain for spatial diversity and low correlationamong antennas for spatial multiplexing. These different andcontradicting requirements require some new antenna designmethods.One potential solution is to introduce the concept of antennagrouping and virtual subarrays [96]. By dividing existingantennas into two or more groups, we can designate tasks ofC&S and optimize the design across groups of antennas. Therecould be overlap between different groups of antennas. Usingvirtual subarrays, we can conveniently generate multibeams[97] satisfying different beamforming requirements from C&S.We can also virtually optimize the antenna placement, by an-tenna selection and grouping. While designing the virtual sub-arrays, we can explore the following commonalities betweenMIMO communication and radar. Similar to the diversity andmultiplexing trade-off in communications, there is a trade-offbetween processing gain and resolution in sensing, related tothe number of independent spatial streams.Considering the benefits of antenna grouping for both C&S,using hybrid antenna arrays [57], [98] will be an attractivelow-cost option. This is particularly true for mmWave systemswhere propagation loss is high and beamforming gain isessential for achieving sufficiently high SNR for both C&S.The research on hybrid array JCAS systems is still in its veryearly stage.
2) Sparse Array Design:
Besides antenna grouping, sparsearray design is another method to exploit the degrees offreedom that can be achieved via configuring the locationsof antennas when the total number of antennas is fixed.Sparse array design, such as coprime array [99], is oftencast as optimally placing a given number of antennas on a larger number of possible uniform grid points [100]. In thisway, a small number of antennas can span a large arrayaperture with a high spatial resolution and low sidelobes.So far, the sparse array design-based JCAS has mainly beenstudied in integrating communication to radar systems, i.e.,embedding information into radar waveforms to perform datacommunication [50], [100]. In [100], antenna position andbeamforming weights are optimized to design beams withmainlobe performing radar detection and sidelobe for commu-nications through modulations like ASK or PSK. In [50], theMIMO waveform orthogonality is further exploited to permutethe waveform across selected antenna grids and hence conveyextra information bits.Sparse array design is particularly suitable for massiveMIMO array with tens to hundreds antennas but a limitednumber of RF chains, i.e., switched arrays or hybrid arrays.This setup can provide more degrees of freedom and potentialperformance enhancement, with reduced cost, in PMNs. Forexample, the sparse array design can add index modulationto the communication part; while the sparse array design canprovide better spatial resolution for radar detection. To thisend, some interesting problems remain to be solved, such ashow to formulate the problems with two goals satisfied andnew trade-offs between C&S.
3) Spatial Modulation:
Spatial modulation uses the setof antenna indexes to modulate information bits and havebeen extensively investigated for communication systems. Formulti-antenna JCAS systems, spatial modulation can alsobe potentially applied. In [41], [46], a concept similar tospatial modulation is exploited to increase communicationdata rate in a frequency-hopping MIMO DFRC system. In[101], spatial modulation is applied to JCAS by allocatingantenna elements based on the transmitted message, achievingincreased communication rates by embedding additional databits in the antenna selection. A prototype is developed in [101]and demonstrates that the proposed scheme can improve theangular resolution and reduce the sidelobe level in the transmitbeam pattern compared to using fixed antenna allocations.Although these works are based on pulsed and continuous-waveform radars, they can potentially be extended to PMN,by adding antenna selection to existing space-time modula-tions. In particular, the rich scattering environment in PMNprovides lower correlation between spatial channels, leadingto potentially better performance.
D. Clutter Suppression Techniques
In PMNs, we treat multipath signals as clutter if they remainlargely unchanged and have near-zero Doppler frequenciesover a period of interest. A lot of clutter could be present inthe received signals because the rich multipath environmentof mobile networks. Clutter contains little information and isbetter to be removed from the signals being sent to the sensingparameter estimator.As discussed in Section V-B, clutter suppression techniquesin traditional radar [1], [74]–[76] may be improved and used inPMNs, but they cannot be directly applied. These techniquestypically need to exploit different features of desired and unwanted echoes, such as low correlation between them. Thesedifferent features may not always be available in mobilenetworks, because the desired multipath and clutter can comefrom the same classes of reflectors.Alternative approaches exploit the correlation in time, fre-quency and space domains, and use recursive averaging ordifferential operation to construct or remove clutter signals[35], [102], [102]–[106]. These approaches could be more vi-able for perceptive mobile networks. They have similarities to background subtraction in image processing [107]. However,there are two major differences: • In image processing, the difference between two imagesis exhibited via pixel variation. In radio sensing, bothDoppler shifts and variation in sensing parameters causedifference in received sensing signals at different time; • In an image, background is overlapped/covered by fore-ground. In radio sensing, clutter and desired multipathsignals are typically additive, and coexist in the receivedsignals.Nevertheless, the many background subtraction methods de-veloped for image processing can be revised and appliedfor radio sensing in PMNs. Below we review two types oftypical background subtraction techniques that can be usedin PMNs: recursive moving averaging (RMA) and Gaussianmixture model (GMM).1) Recursive Moving Averaging (RMA):
Assume sensingparameters are fixed over the coherence time period, thenideally the received signals for each path at two differenttimes will only have a phase difference caused by the Dopplerphase shift. If the Doppler frequency is near zero, then the twosignals are nearly identical. Based on this assumption, we canuse use an RMA method [35] to estimate the clutter and thenremove it from the received signal.The RMA method uses a small forgetting vector to recur-sively average the received signal over a window, with a lengthsufficiently large to allow suppressing time-varying signalsof non-static paths, but smaller than the coherent time. Thewindow length can be adapted to the variation speed of thechannels. The time interval between the inputs to the averagingdetermines how signals with different Doppler frequencies areadded, either constructively or destructively. Hence it has asignificant impact on suppressing signals of different Dopplerfrequencies. The forgetting factor and the window lengthdetermine the suppression power ratio. Although experimentalresults have been reported in [35] for the relationship betweenthese parameters and the effect of clutter estimation andsuppression, optimal combinations of these parameters, inconsideration of channel statistical properties, are yet to bestudied.Although the RMA method works well in principal, it maybecome inefficient due to practical issues, such as timing andfrequency offset commonly existing in actual systems. Thesesignal imperfectness needs to be well compensated before theRMA method can work effectively.
2) Gaussian Mixture Model:
GMM has been widely usedfor analyzing and separating moving objects from the back-ground in image and video analysis [107], target identification and classification in radar system [108], and positioning solu-tions [109]. The statistical learning of the GMM model withrespect to the mean and variance in background subtraction isused to determine the state of each pixel whether a pixel isbackground or foreground. It has also been applied recentlyto extract static channel state information from channel mea-surement in [110]. Different from GMM in video analysiswhere background and foreground overlap each other, clutterand multipath of interest in PMNs are additive and cancoexist. Therefore, it is infeasible in PMNs to place foreground(dynamic signals) and background (static signals) into twodifferent sets by classical clustering approaches that happenedin image or video signal processing.GMM’s working principle for clutter suppression in PMNsis as follows. Wireless channels can be modeled and estimatedby a mixture of Gaussian distributions since each densityrepresents multipaths in the channel [110]. Static and dynamicpaths can be represented by Gaussian distributions with verydifferent parameters over the time domain. This is becauseover a short time period, static paths change little and dynamicpaths may vary significantly. It is also quite common thatstatic paths typically have larger mean power than dynamicones. Hence, in terms of their distributions, static paths havenear-zero variances, which are much smaller than those of thedynamic ones. Therefore, by learning the mean values of thedistribution, static paths can be identified and separated viacomparing the variance.The main advantage of GMM for clutter estimation in PMMis that much less samples are required to achieve a givenaccuracy, compared to the matched filtering and RMA meth-ods. However, the estimation usually needs to be realized byhigh-complexity algorithms such as expectation maximization.Low-complexity estimation based on the GMM formulation isa key research problem here.Fig. 5 compares the root mean square error (RMSE) resultsfor clutter estimation between RMA and GMM methods. Thesignal to interference ratio Υ denotes the ratio between clutter-to-dynamic power ratio. The estimation for GMM is based on samples. For RMA, the forgetting factor is . over and samples. According to the figure, the GMM methodachieves significantly lower RMSE for clutter estimation thanRMA at both r = 10 and iterations. E. Sensing Parameter Estimation
The tasks of sensing in PMNs include both explicit es-timation of sensing parameters for locating objects and es-timating their moving speeds, and application oriented pat-tern recognition such as object and behaviour recognitionand classification. In this subsection, we review research onsensing parameter estimation, considering typical multiuser-MIMO OFDM signals used in modern mobile networks. Wewill review work on pattern recognition in subsection VI-F.We note that sensing parameter estimation is a non-linearproblem, and hence most classical linear estimators, whichhave been widely used in channel estimation in communi-cations, cannot be applied. Here, we review the followingtechniques: periodogram such as 2D DFT, subspace based
Signal to Interference Ratio (dB), Υ -6 -5 -4 -3 -2 C l u tt e r e s t i m a t i on R M SE ( d B ) RMA @ r=10RMA @ r=150 GMM-EM-CE @ N m =10 Fig. 5:
Simulation results comparing different clutter suppressionmethods. spectrum analysis techniques, on-grid compressive sensing(CS) algorithms, off-grid CS algorithms and grid densification,Tensor tools, estimation in clustered channels, and resolutionof sensing ambiguity. Most of these techniques have highercomplexity than classical channel estimation algorithms. Sincethe required sensing rate is typically at the order of mil-liseconds to seconds, such high computational complexity isaffordable at BSs. Comparison of some of these techniquesfor sensing parameter estimation in PMNs is summarized inTable VI. Details of the research are elaborated as follows.
1) Periodogram such as 2D DFT:
The classical 2D DFTmethod is a periodogram method being widely used in radar.It can be used to coarsely estimate sensing parameters bycombining two of the following three transformations: con-verting the time-domain samples to frequency domain, spatial-domain samples to angle domain, and phase shifting samplesto Doppler frequency domain. A 3D DFT may also be used.But due to the complexity, it is generally replaced by two orthree 2D DFTs. The resolution of this method is low becauseof the long tail of the inherent sinc function in the DFT.A windowing operation can be applied to slightly improvethe resolution. This method typically requires a full set ofcontinuous measurements in time or frequency domain, whichcan limit its application in PMNs due to the discontinuoussamples.
2) Subspace Based Spectrum Analysis Techniques:
Clas-sical subspace based spectrum analysis techniques such asMUSIC and ESPRIT can estimate parameters of continuousvalues with high resolution [111]. However, their applica-tions in PMNs may also be limited as they typically requiresamples of equal intervals. Techniques that can deal withnon-uniform sampling have been proposed, e.g., the coupledcanonical polyadic decomposition approach in [112] and thegeneralized array manifold separation approach in [113], butthey have very high computational complexity. To achievehigh resolutions, MUSIC and ESPRIT typically also requirea large number of samples so that the signal subspace andnoise subspace can be well separated. This may not always beavailable in some domains, such as the spatial domain, which TABLE VI: Comparison of Sensing Parameter Estimation Algorithms
Algorithms Properties Suitability and main limitation
Periodogram suchas 2D DFT Simple, but low resolution. May be used as thestarting point for other algorithms. Generally, requires a full set of continuous samples inall domains, which may not always be satisfied.Subspace methodssuch as ESPRITand MUSIC High resolution and can do off-grid estima-tion. High complexity. High dimension Tensorbased ESPRIT and MUSIC algorithms withreduced complexity are also available. Typically require a large segment of consecutive sam-ples, which may not always be satisfied.Compressive sens-ing (On-grid) Flexible. Does not require consecutive sam-ples. Various recovery algorithms that can beselected to adapt to complexity and perfor-mance requirements. Works well even for estimating a small amount of off-grid parameters. Performance can degrade significantlywith many paths of continuous parameter values.CompressiveSensing (Off-grid)such as atomicnorm minimization Flexible and do not require consecutive sam-ples. Capable of estimating off-grid values. Limitation in real time operation due to very highcomplexity. Still require sufficient separation betweenparameter values.Tensor based algo-rithms High-order formulation using the Tensor toolssuch as 3D Tensor CS simplified computationalcomplexity and provides capability in resolvingmultipath with repeated parameter values. Tensor tools need to be combined with other algo-rithms such as ESPRIT and CS. Thus they face theinherent problems of these algorithms. would require a large number of antennas. However, it maybe a good option to combine them with other techniques forsensing parameter estimation, by exploiting their capabilitiesof high resolution and estimating parameters of continuousvalues.
3) On-Grid Compressive Sensing Algorithms:
CompressiveSensing (CS) techniques [114] have been widely used incommunication systems for channel estimation [115]–[119]and in radar systems [120]. CS techniques formulate parameterestimation as a sparse signal recovery problem, which can besolved by many algorithms such as l recovery (convex relax-ation), greedy algorithms and probabilistic inference [114]. Atthe least, only twice the number of samples are required toaccurately recover a certain number of unknown parameters,in the noise free case. Typical CS techniques use on-gridquantized dictionaries, and hence errors are caused due toquantization when the original parameters have continuousvalues. One main advantage of CS for sensing parameterestimation in PMNs is that it does not require consecutivesamples. Actually, higher randomness of samples in time,frequency and spatial domains can generally lead to betterestimation performance.The sensing parameters to be estimated in PMNs include de-lay, AoA and Doppler in three different domains. Sometimes,the angle of departure (AoD) and magnitude of path are alsoof interest, which are not considered here. Since the signalsare relatively independent in the three domains, they can beformulated in a high-dimension (3D here) vector Kroneckerproduct form or even Tensor form. Therefore, we can apply 1Dto 3D CS techniques to estimate these sensing parameters. Thefollowing two problems need to be considered when selectingCS techniques of different dimensions. • Quantization error and number of available samples:
Al-though high-dimensional on-grid CS algorithms such asthe Kronecker CS [121] could offer better performance, they require more samples than unknown variables ineach dimension. In a typical BS, we can get sufficientnumber of observations for the delay (linked to subcar-riers), a reasonable number of samples in the Dopplerfrequency domain (linked to intermittent packets overa segment of channel coherent period), and a limitednumber of AoA observations (linked to antennas). • Complexity:
Exploiting the Kronecker CS property, thecomputational complexity is in the order of the productof the complexity in each domain, which is typicallyproportional to the cube of the number of samples.Therefore, a high dimensional CS algorithm is not alwaysthe viable option, particularly for the Doppler frequencyand AoA estimation due to the limited number of sam-ples. Comparatively, mobile signals generally have tens tothousands of subcarriers, which provide numerous samplesfor delay estimation. Thus, we can formulate two multi-measurement vector (MMV) CS problems, by stacking spatial-domain and Doppler-frequency domain signals, respectivelywith frequency domain signals. From the MMV-CS amplitudeestimates, we can then estimate the AoA and Doppler frequen-cies [35], [39]. The details of CS algorithms from 1D to 3Dand their performance are presented in [39]. One commonproblem associated with using lower dimension CS is thatparameters with overlapped values in one or more dimensionscannot be separately estimated. In this case, techniques suchas the one proposed in [122] can be used, by taking advantageof the capability of model-based algorithms, for example,modified matrix enhancement and matrix pencil.For multiuser-MIMO signals, for example, signals receivedat an RRU from multiple RRUs in downlink passive sensing,we can use two methods to formulate the CS problems [35].The first, direct sensing method, directly uses the receivedsignals as inputs to CS sensing algorithms. Since the receiverknows the transmitted information data symbols, the problem can be formulated as a block CS model [35], [118], [123],without decorrelating signals from multiusers. Correlationbetween the parameters can also be exploited in this model, viaintroducing intra-block correlation coefficients. The second, indirect sensing , is based on signal stripping that decorrelatessignals between users [35], [37]. Then the sensing parameterscan be estimated for each individual user by conventional CSalgorithms. Direct sensing can achieve better performance thanindirect sensing, as the decorrelation process introduces noiseenhancement, at the cost of higher complexity. If the datasymbols are unknown, e.g., in uplink sensing, decorrelatingand demodulation errors also exist. Such errors may be explic-itly considered and removed in the estimation [124]. In [124],a passive sensing algorithm for multiple objects is proposedby using demodulated signals. The delay-Doppler values areestimated by exploiting the sparsity of the demodulation errorsand numbers of objects. The positions and velocities of objectsare then estimated based on the estimated delay-Doppler, usingneural network techniques.Overall, on-grid CS algorithms are promising for sensingparameter estimation in PMNs. However, the quantizationerror is a major problem as true sensing parameters havecontinuous values. For parameters of continuous values, thereexist mismatch between the assumed and actual dictionaries,generally known as “dictionary mismatch”, which can causesignificant performance degradation [125]. The degradationis severer when the number of unknown variables is larger.Therefore, resolving the quantization error and dictionarymismatch is a major challenge here.
4) Grid Densification and Off-Grid CS Algorithms:
Thereare mainly two types of techniques that have been developed totackle the quantization error problem in CS: grid densificationand off-grid CS algorithms [126], [127]. Both techniques havehigher complexity than conventional on-grid CS algorithms.Grid densification uses denser dictionaries to reduce quan-tization error. The discretization of the physical space isunavoidable since CS has been focused on the signals thatcan be represented under a finite dictionary by reconstruction.It is intuitively reasonable that both dictionary mismatch andparameter estimation error can be reduced with a dense grid.Therefore, the question comes whether a denser grid leads tomore accurate sparse signal recovery or not. In fact, accordingto the CS theory, the sampled grids should not be too dense.As in densely sampled grids, the dictionaries have a high inter-column correlation. The high correlation of dictionary itemsviolates the restricted isometry property (RIP) condition of CS[115]. This is particularly of concern when the SNR is not veryhigh. Therefore, there is a trade-off in dictionary mismatchand estimation accuracy while constructing a densified dic-tionary. Dynamic dictionaries with multi-resolution capabilityare proposed to resolve this problem. For example, in [128], adynamic dictionary based re-focused DOA estimation methodis developed with the number of extremely sparse grids refinedto the number of detected sources.There are extensive research interests in extending CS to off-grid models, via, e.g., the perturbation method [129], CS plusmaximal likelihood [130], and the atomic norm minimization(ANM) method [72], [126], [131]. The ANM method [126], [131] can handle continuous dictionary and recover unknownvariables with a reasonable number of samples at a highprobability via a semidefinite program. It has been widelyapplied for channel estimation in, e.g., generalized spatialmodulation systems [72], MIMO radar via MMV models[132], and mmWave MIMO systems with planar arrays [133].However, the ANM method still requires that the variablessuch as delays have well separated values. This may notalways be satisfied in PMNs as an object may not always beapproximated as a point reflector/scatter and reflected/scatteredsignals may come in clusters due to the limited distance amongthe transmitter, the object and the receiver. Enhancing theANM method and making it capable of handling such signalsare important for its practical application in PMNs.
5) Sensing with Clustered Multipath Channels:
In clustersparsity patterns, non-zero taps of sparse signal appear in clus-ters rather than being arbitrarily spread over the vector, whichmeans that sparse signal exhibits a structure in the form of non-zero coefficients occurring in clusters. In practice, multipathsignals in mobile systems often arrive in clusters [134], andpaths from one cluster typically come from the same scatter(s)and have similar parameter values. The situation becomescomplex once the clusters originated in a propagation scenehave correlation among other clusters of the same user andacross different users. Eventually getting sensing parametersfrom delay or spatial domain without acknowledging thechannel cluster structure can create accuracy problems.We can find several research results on reconstructing clus-ter sparse signals in general, for example, through periodiccompressive support [135], model based CS [136], variationalBayes approach [137], and block Bayesian method [138]. Theexploitation of the cluster property in multipath channels forsensing parameter estimation in PMNs is possible throughcreating a prior probability distribution. In particular, a clusterprior probability density function needs to be introduced inthe CS reconstruction algorithm in order to efficiently detectthe coarse locations of the clusters, leading to more accuratesparse reconstruction performance when CS algorithms areapplied. Detailed technology on how cluster sparsity canbe exploited in JCAS systems such as PMNs that involveOFDMA and multi-user MIMO is yet to be developed.
6) Resolution of Sensing Ambiguity:
As discussed in Sec-tion IV, there is typically no clock-level synchronization be-tween a sensing receiver and the transmitter in PMNs, particu-larly in uplink sensing. In this case, there exist both timing andcarrier frequency offsets in the received signals. The timingoffset is typically time-varying, i.e., it has a random valuewhich can change during any two discontinuous transmission.The carrier frequency offset (CFO) may slowly vary over timedue to oscillator stability. Unlike the case in communications,where timing offset and CFO can be absorbed into channelestimation, in sensing they cause measurement ambiguity andaccuracy degradation. Timing offset can directly cause timingambiguity and then ranging ambiguity, and CFO can causeDoppler estimation ambiguity and then speed ambiguity. Theyalso prevent aggregating signals from discontinuous packetsfor joint processing, as they cause unknown and different phaseshifting across packets. There have been a limited number of works that addressthis problem in passive sensing [139]–[141]. A cross-antennacross-correlation (CACC) method is applied to passive WiFi-sensing, to resolve the timing ambiguity issues. The basicassumption is that timing offsets across multiple antennas inthe receiver are the same, and hence they can be removedby computing the cross-correlation between signals from mul-tiple receiving antennas. In [141], CACC is used to obtainestimates for ranges and velocities of targets. In [140], CACCis adopted to get the angle-of-arrival (AoA) spectrum, whichrepresents the probabilities of the direction or angle of target.However, the outputs after CACC contain cross-product termsand actually doubled the number of unknown parameters tobe estimated. The authors in [140] proposed a method tosuppress signals containing half of the unknown parameters,but the method is found to be susceptible to the number andpower distribution of static and dynamic signal propagationpaths. Therefore, although the idea of CACC looks attractivein resolving the sensing ambiguity problem, more advancedtechniques need to be developed to handle the output signalsfrom CACC. In PMNs, the transmitted signals may alsobe optimized to enable better implementation of the CACCmethod.
F. Pattern Analysis
Using radio signals, high-level application-oriented object,behaviour and event recognition and classification can beachieved by combining machine learning and signal processingtechniques. They can be realized with or without using thesensing parameter estimation results, which provide locationand velocity information.The feasibility and benefits of applying machine learn-ing technologies to communication systems have been welldemonstrated, for example, fast beamforming design via deeplearning [142], behavioral modeling and linearization of wide-band RF power amplifiers in 5G system [143], vehicularnetwork modeling in 6G by machine learning [144], routecomputation for software defined communication systems bydeep learning strategy [145], and heterogeneous network trafficcontrol by deep learning [146].Although the work on pattern analysis using mobile signalsis still in its infancy stage, we have seen some interestingexamples, such as [23]–[25]. We can foresee its booming in thenear future, as we have been observing from many successfulWiFi sensing applications. Using WiFi signals for object andbehaviour recognition and classification has been well demon-strated [147]–[151]. Mobile signals are more complicatedthan WiFi signals, and the outdoor propagation environmentis also more challenging. However, the PMNs have moreadvanced infrastructure than WiFi systems, including largerantenna arrays, more powerful signal processing capability,and distributed and cooperative nodes. Using massive MIMO,a PMN BS equivalently possesses a massive number of“pixels” for sensing. It is able to resolve numerous objectsat a time and achieve imaging results with better field-of-viewand resolution, like optical cameras.Based on the various approaches developed for WiFi sens-ing, we can deduce the procedures of applying pattern anal-
Signal Collection (Rate) Signal Preprocessing(Stripping, Cleaning, Compression) Feature Extraction (Supervised, non‐supervised) Recognition and ClassificationSensing parameter estimation
Fig. 6: Block diagram showing the procedure for patternrecognition.ysis to mobile signals, as shown in Fig. 6. They typicallyinvolve four steps: signal collection, signal preprocessing,feature extraction, and recognition and classification. In thesignal collection step, the signals are collected at the receiveraccording to the desired rate. In the signal preprocessing step,the collected signals may be stripped, cleaned and compressed.Signal stripping removes the modulated symbols from thereceived signal, and hence the pure channel state information(CSI) is obtained. Multiuser signals may also be decorrelatedhere. Signal cleaning removes signal distortions associatedwith, e.g., timing, CFO and phase noise, and suppressesclutter signals. The purpose is to keep mostly informationcarrying signals. Many of the algorithms described beforecan be applied for this purpose. If signals arrive irregularlyin the first step, the CSI can also be interpolated here ifdesired. Signal compression makes the signal condense, sothat the useful information can be enhanced and the processingcomplexity in the following steps can be reduced. Commoncompression techniques include principal component analysis(PCA) and correlation [151]. Feature signals are then extractedfrom preprocessed signals, using machine learning techniquessuch as supervised and non-supervised deep learning. Finally,recognition and classification are conducted, with inputs fromthe extracted feature signals, the preprocessed signals, andestimated sensing parameters.
G. Networked Sensing under Cellular Topology
PMNs provide great opportunities for radio sensing undera cellular structure, which could be well beyond the scale andcomplexity of distributed radar systems. The main challengefor networked sensing under a cellular topology remains in theway to address competition and cooperation between differentnodes for sensing performance characterization and algorithmdevelopment. The research in this area is almost blank at themoment. Here, we envision two potential research directions.
1) Fundamental Theories and Performance Bounds for“Cellular Sensing Networks”:
This is about investigating thepotentials of the cellular structure on improving the spec-tral efficiency and performance of sensing, and developingfundamental theories and performance bounds for such im-provement. Similar to communications, a cellular networkintuitively also increases frequency reuse factor and hencethe overall capacity for sensing. Stochastic geometry modelmay be an excellent tool for analyzing the dynamics in thesensing network, as have been applied to characterize theaggregated radar interference in an autonomous vehicular net- work in [152], [153]. Both intra-cell and inter-cell interferencewould then be taken into consideration in deriving the mutualinformation for networked sensing.
2) Distributed Sensing with Node Grouping and Coopera-tion:
One way of exploiting networked sensing is to developdistributed and cooperative sensing techniques by schedulingand grouping UEs and enable cooperation between RRUs. Onone hand, existing research has shown that distributed radartechniques can improve location resolution and moving targetdetection by providing large spatial diversity and wide angularobservation [154]. Such diversity can be maximized by opti-mizing both waveform design and placement of radar nodes. InPMNs, we can group multiple UEs sensing results to improveuplink sensing. On the other hand, distributed radar canenable high-resolution localisation, exploiting coherent phasedifference of carrier signals from different distributed nodes[82]. This requires phase synchronisation among radar nodes,and can only be potentially achieved in downlink sensing bygrouping RRUs. For both cases, we may develop distributedsensing techniques, leveraging on extensive research works ondistributed beamforming and cooperative communications.
H. Sensing-Assisted Secure Communication
When communication and sensing are integrated, it isimportant to understand how they can mutually benefit fromeach other. Existing research has investigated how to use thechannel structure obtained in sensing to improve the reliabilityof communications [155], [156]. Such detailed information onchannel composition may play a more important role in securewireless communication, with the application of physical layersecurity techniques. Current physical layer security studies aremainly based on channel state information. Comparatively, thesensing results contain more essential information about theenvironment between a pair of transmitter and receiver. Theycan motivate more informative secret-key generating methodsand agreement in cellular communication networks. As a start,we can characterize the secrecy capacity of PMNs, and developpractical code design methods for information encryption.VII. C
ONCLUSION
We have provided a comprehensive review on the perceptivemobile network (PMN), which integrates radio sensing intocurrent communication-only mobile network, using the jointcommunication and radio/radar sensing (JCAS) techniques.Referring to 5G NR standard, we have illustrated that uplinkand downlink sensing can be realized with different degreesof modifications and enhancement to current mobile networkinfrastructure. We have provided a detailed review for majorresearch challenges, potential solutions and diverse researchopportunities within the context of PMN.The PMN is expected to deliver a revolutionary ubiquitousradio sensing network that can significantly drive smart ini-tiatives such as smart cities and smart transportation, inte-grated with enriched mobile communication. In relation to the(stereo) optical vision in camera sensing, the PMN is expectedto realise 3D+ radio vision, including 3D location + speed +features for objects surrounding the radio transceivers, with additional attractive features such as day-and-night availability,fog/leaf-penetration, and continuous tracking. It will enablemany new applications for which current sensing solutions areimpractical or too costly. While there are significant challengesand a long way ahead to make the PMN fully operational, oursurvey here is a solid presentation, indicating the feasibilityand providing the potential directions to pursue.R
EFERENCES[1] C. Sturm and W. Wiesbeck, “Waveform Design and Signal ProcessingAspects for Fusion of Wireless Communications and Radar Sensing,”
Proceedings of the IEEE , vol. 99, no. 7, pp. 1236–1259, 2011.[2] L. Han and K. Wu, “Joint wireless communication and radar sensingsystems state of the art and future prospects,”
IET Microwaves,Antennas and Propagation , vol. 7, no. 11, pp. 876–885, 2013.[3] B. Paul, A. R. Chiriyath, and D. W. Bliss, “Survey of RF communi-cations and sensing convergence research,”
IEEE Access , vol. 5, pp.252–270, 2017.[4] L. Zheng, M. Lops, Y. C. Eldar, and X. Wang, “Radar and communi-cation coexistence: An overview: A review of recent methods,”
IEEESignal Processing Magazine , vol. 36, no. 5, pp. 85–99, Sep. 2019.[5] A. R. Chiriyath, B. Paul, and D. W. Bliss, “Radar-CommunicationsConvergence: Coexistence, Cooperation, and Co-Design,”
IEEE Trans-actions on Cognitive Communications and Networking , vol. 3, no. 1,pp. 1–12, 2017.[6] A. Gameiro, D. Castanheira, J. Sanson, and P. Monteiro, “Researchchallenges, trends and applications for future joint radar communica-tions systems,”
Wireless Personal Communications , vol. 100, no. 1,March 2018.[7] E. Cianca, M. D. Sanctis, and S. D. Domenico, “Radios as sensors,”
IEEE Internet of Things Journal , vol. 4, no. 2, pp. 363–373, April2017.[8] F. Liu, C. Masouros, A. Petropulu, H. Griffiths, and L. Hanzo, “Jointradar and communication design: Applications, state-of-the-art, and theroad ahead,”
IEEE Transactions on Communications , pp. 1–1, 2020.[9] J. A. Zhang, X. Huang, Y. J. Guo, J. Yuan, and R. W. Heath,“Multibeam for joint communication and radar sensing using steerableanalog antenna arrays,”
IEEE Transactions on Vehicular Technology ,vol. 68, no. 1, pp. 671–685, Jan 2019.[10] F. Liu, L. Zhou, C. Masouros, A. Li, W. Luo, and A. Petropulu, “To-ward dual-functional radar-communication systems: Optimal waveformdesign,”
IEEE Transactions on Signal Processing , vol. 66, no. 16, pp.4264–4279, Aug 2018.[11] F. Liu, C. Masouros, A. Li, H. Sun, and L. Hanzo, “MU-MIMOcommunications with MIMO Radar: From co-existence to joint trans-mission,”
IEEE Transactions on Wireless Communications , vol. 17,no. 4, pp. 2755–2770, April 2018.[12] A. F. Martone, K. A. Gallagher, and K. D. Sherbondy, “Joint radarand communication system optimization for spectrum sharing,” in , April 2019, pp. 1–6.[13] R. Saruthirathanaworakun, J. M. Peha, and L. M. Correia, “Oppor-tunistic sharing between rotating radar and cellular,”
IEEE Journal onSelected Areas in Communications , vol. 30, no. 10, pp. 1900–1910,November 2012.[14] M. Labib, V. Marojevic, A. F. Martone, J. H. Reed, and A. I. Zaghloui,“Coexistence between communications and radar systems: A survey,”
URSI Radio Science Bulletin , vol. 2017, no. 362, pp. 74–82, Sep. 2017.[15] E. Yousif, F. Khan, T. Ratnarajah, and M. Sellathurai, “On the spectralcoexistence of colocated MIMO radars and wireless communicationssystems,” in , July 2016,pp. 1–5.[16] F. Liu, C. Masouros, A. Li, and T. Ratnarajah, “Robust MIMObeamforming for cellular and radar coexistence,”
IEEE Wireless Com-munications Letters , vol. 6, no. 3, pp. 374–377, June 2017.[17] B. Friedlander, “Waveform design for MIMO radars,”
IEEE Transac-tions on Aerospace and Electronic Systems , vol. 43, no. 3, pp. 1227–1238, July 2007.[18] D. E. Hack, L. K. Patton, B. Himed, and M. A. Saville, “Detectionin passive MIMO radar networks,”
IEEE Transactions on SignalProcessing , vol. 62, no. 11, pp. 2999–3012, June 2014.[19] S. Gogineni, M. Rangaswamy, B. D. Rigling, and A. Nehorai, “Cramr-Rao bounds for UMTS-based passive multistatic radar,”
IEEE Trans-actions on Signal Processing , vol. 62, no. 1, pp. 95–106, Jan 2014. [20] R. S. A. R. Abdullah, A. A. Salah, A. Ismail, F. Hashim, N. E. A.Rashid, and N. H. A. Aziz, “LTE-based passive bistatic radar systemfor detection of ground-moving targets,” ETRI Journal , vol. 38, no. 2,pp. 302–313, 2016.[21] C. R. Berger, B. Demissie, J. Heckenbach, P. Willett, and S. Zhou,“Signal processing for passive radar using OFDM waveforms,”
IEEEJournal of Selected Topics in Signal Processing , vol. 4, no. 1, pp. 226–238, Feb 2010.[22] K. Chetty, G. E. Smith, and K. Woodbridge, “Through-the-WallSensing of Personnel Using Passive Bistatic WiFi Radar at StandoffDistances,”
IEEE Transactions on Geoscience and Remote Sensing ,vol. 50, no. 4, pp. 1218–1226, April 2012.[23] R. Harris, J.-K. Lam, and E. F. Burroughs, “The potential of cell phoneradar as a tool for transport applications,” Association for EuropeanTransport and contributors, Report, 2005.[24] N. David, O. Sendik, H. Messer, and P. Alpert, “Cellular Network In-frastructure: The Future of Fog Monitoring?”
Bulletin of the AmericanMeteorological Society , vol. 96, no. 10, pp. 1687–1698, 2015.[25] R. Uijlenhoet, A. Overeem, and H. Leijnse, “Opportunistic remotesensing of rainfall using microwave links from cellular communicationnetworks,”
WIREs Water , vol. 5, no. 4, p. e1289, 2018.[26] M. A. A. Al-qaness, “Device-free human micro-activity recognitionmethod using wifi signals,”
Geo-spatial Information Science , vol. 22,no. 2, pp. 128–137, 2019.[27] J. Liu, H. Liu, Y. Chen, Y. Wang, and C. Wang, “Wireless sensing forhuman activity: A survey,”
IEEE Communications Surveys Tutorials ,pp. 1–1, 2019.[28] S. Wang and G. Zhou, “A review on radio based activity recognition,”
Digital Communications and Networks , vol. 9, 03 2015.[29] J. A. Zhang, A. Cantoni, X. Huang, Y. J. Guo, and R. W. H. Jr,“Framework for an Innovative Perceptive Mobile Network Using JointCommunication and Sensing,” in . IEEE, June 2017, Conference Proceedings,pp. 1–5.[30] J. A. Zhang, X. Huang, Y. J. Guo, and R. W. H. Jr, “Joint Communi-cations and Sensing Using Two Steerable Analog Antenna Arrays,” in , June2017, Conference Proceedings, pp. 1–5.[31] P. Kumari, J. Choi, N. G. Prelcic, and R. W. Heath, “IEEE 802.11ad-based Radar: An Approach to Joint Vehicular Communication-RadarSystem,”
IEEE Transactions on Vehicular Technology , vol. PP, no. 99,pp. 1–1, 2017.[32] R. C. Daniels, E. R. Yeh, and R. W. Heath, “Forward collisionvehicular radar with IEEE 802.11: Feasibility demonstration throughmeasurements,”
IEEE Transactions on Vehicular Technology , vol. 67,no. 2, pp. 1404–1416, Feb 2018.[33] Y. Luo, J. A. Zhang, X. Huang, W. Ni, and J. Pan, “Optimizationand quantization of multibeam beamforming vector for joint commu-nication and radio sensing,”
IEEE Transactions on Communications ,vol. 67, no. 9, pp. 6468–6482, Sep. 2019.[34] V. Petrov, G. Fodor, J. Kokkoniemi, D. Moltchanov, J. Lehtomaki,S. Andreev, Y. Koucheryavy, M. Juntti, and M. Valkama, “On unifiedvehicular communications and radar sensing in millimeter-wave andlow terahertz bands,”
IEEE Wireless Communications , vol. 26, no. 3,pp. 146–153, June 2019.[35] M. L. Rahman, J. A. Zhang, X. Huang, Y. J. Guo, and R. W.Heath Jr, “Framework for a perceptive mobile network using jointcommunication and radar sensing,”
IEEE Transactions on Aerospaceand Electronic Systems , pp. 1–1, 2019.[36] R. M. Gutierrez, H. Yu, A. R. Chiriyath, G. Gubash, A. Herschfelt, andD. W. Bliss, “Joint sensing and communications multiple-access systemdesign and experimental characterization,” in , March 2019, pp. 1–8.[37] J. A. Zhang, X. Huang, Y. J. G. and, and M. L. Rahman, “SignalStripping Based Sensing Parameter Estimation in Perceptive MobileNetworks,” in . IEEE, 2017,Conference Proceedings, pp. 67–70.[38] M. L. Rahman, J. A. Zhang, X. Huang, and Y. J. Quo, “Analog antennaarray based sensing in perceptive mobile networks,” in , Sep. 2017, pp. 199–202.[39] M. L. Rahman, P. Cui, J. A. Zhang, X. Huang, Y. J. Guo, andZ. Lu, “Joint communication and radar sensing in 5G mobile networkby compressive sensing,” in , Sep. 2019, pp.599–604. [40] J. Li and P. Stoica. John Wiley & Sons, Ltd, 2008.[41] A. Hassanien, M. G. Amin, E. Aboutanios, and B. Himed,“Dual-function radar communication systems: A solution to thespectrum congestion problem,”
IEEE Signal Processing Magazine ,vol. 36, no. 5, pp. 115–126, sep 2019. [Online]. Available:https://doi.org/10.1109%2Fmsp.2019.2900571[42] C. W. Rossler, E. Ertin, and R. L. Moses, “A software defined radarsystem for joint communication and sensing,” in . IEEE, 2011, Conference Proceedings, pp.1050–1055.[43] B. Paul, “RF Convergence of Radar and Communications: Metrics,Bounds, and Systems,” PhD Thesis, Electrical engineering, 2017.[44] H. Lou, Q. Zhang, B. Liang, and Y. Chen, “Waveform design andanalysis for radar and communication integration system,” in
IETInternational Radar Conference 2015 , Oct 2015, pp. 1–6.[45] Y. Zhang, Q. Li, L. Huang, Song, and J., “Waveform Design ForJoint Radar-Communication System With Multi-User Based On MIMORadar,” in , 2017, Confer-ence Proceedings, pp. 0415–0418.[46] K. Wu, J. Guo, X. Huang, and R. W. H. Jr., “Accurate Channel Esti-mation for Frequency-hopping Dual-Function Radar-Communication,”in . IEEE, Conference Proceedings, pp. 1–6.[47] Y. L. Sit, C. Sturm, and T. Zwick, “Doppler estimation in an ofdmjoint radar and communication system,” in , March 2011, pp. 1–4.[48] D. H. N. Nguyen and R. W. Heath, “Delay and doppler processing formulti-target detection with ieee 802.11 ofdm signaling,” in , March 2017, pp. 3414–3418.[49] M. Braun, “OFDM radar algorithms in mobile communication net-works,”
PhD thesis, Institut fur Nachrichtentechnik des KarlsruherInstituts fur Technologie, Karlsruhe , 2014.[50] X. Wang, A. Hassanien, and M. G. Amin, “Dual-function MIMOradar communications system design via sparse array optimization,”
IEEE Trans. Aerospace and Electronic Systems , vol. 55, no. 3, pp.1213–1226, 2019. [Online]. Available: https://doi.org/10.1109/TAES.2018.2866038[51] X. Wang, A. Hassanien, and M. G. Amin, “Sparse transmit array designfor dual-function radar communications by antenna selection,”
DigitalSignal Processing , vol. 83, pp. 223–234, 2018.[52] Y. Liu, G. Liao, J. Xu, Z. Yang, and Y. Zhang, “Adaptive ofdmintegrated radar and communications waveform design based on in-formation theory,”
IEEE Communications Letters , vol. 21, no. 10, pp.2174–2177, 2017.[53] F. Liu, C. Masouros, A. Li, H. Sun, and L. Hanzo, “MU-MIMOcommunications with MIMO radar: From co-existence to joint trans-mission,”
IEEE Transactions on Wireless Communications , vol. 17,no. 4, pp. 2755–2770, April 2018.[54] Y. Rong, A. R. Chiriyath, and D. W. Bliss, “Multiple-antenna multiple-access joint radar and communications systems performance bounds,”in ,Oct 2017, pp. 1296–1300.[55] M. Bic and V. Koivunen, “Multicarrier radar-communications wave-form design for rf convergence and coexistence,” in
ICASSP 2019 -2019 IEEE International Conference on Acoustics, Speech and SignalProcessing (ICASSP) , May 2019, pp. 7780–7784.[56] M. Alloulah and H. Huang, “Future millimeter-wave indoor systems:A blueprint for joint communication and sensing,”
Computer , vol. 52,no. 7, pp. 16–24, July 2019.[57] F. Liu and C. Masouros, “Hybrid beamforming with sub-arrayedMIMO radar: Enabling joint sensing and communication at mmwaveband,” in
ICASSP 2019 - 2019 IEEE International Conference onAcoustics, Speech and Signal Processing (ICASSP) , May 2019, pp.7770–7774.[58] K. V. Mishra, M. R. Bhavani Shankar, V. Koivunen, B. Ottersten, andS. A. Vorobyov, “Toward millimeter-wave joint radar communications:A signal processing perspective,”
IEEE Signal Processing Magazine ,vol. 36, no. 5, pp. 100–114, Sep. 2019.[59] P. Kumari, S. A. Vorobyov, and R. W. H. Jr, “Adaptive virtual waveformdesign for millimeter-wave joint communication-radar,” 2019.[60] S. H. Dokhanchi, B. S. Mysore, K. V. Mishra, and B. Ottersten,“A mmwave automotive joint radar-communications system,”
IEEETransactions on Aerospace and Electronic Systems , vol. 55, no. 3, pp.1241–1260, June 2019.[61] Z. Geng, R. Xu, H. Deng, and B. Himed, “Fusion of radar sensing andwireless communications by embedding communication signals into the radar transmit waveform,” IET Radar, Sonar Navigation , vol. 12,no. 6, pp. 632–640, 2018.[62] S. Sur, X. Zhang, P. Ramanathan, and R. Chandra, “BeamSpy: EnablingRobust 60 GHz Links Under Blockage,” in
NSDI’16 Proceedingsof the 13th Usenix Conference on Networked Systems Design andImplementation . USENIX Association Berkeley, CA, USA, 2016,Conference Proceedings, pp. 193–206.[63] S. Sur, V. Venkateswaran, X. Zhang, and P. Ramanathan, “60 GHzIndoor Networking through Flexible Beams: A Link-Level Profiling,”in
SIGMETRICS15 . ACM, 2015, Conference Proceedings, pp. 71–84.[64] J. Lee, H. Noh, and H. J. Yang, “Performance analysis of beamformingfor radar and communications coexisting systems,” in
MobiCom16 . ACM, 2016, Conference Proceedings, pp.162–175.[67] Y. Liu, G. Liao, J. Xu, Z. Yang, and Y. Zhang, “Adaptive OFDMintegrated radar and communications waveform design based on in-formation theory,”
IEEE Communications Letters , vol. 21, no. 10, pp.2174–2177, Oct 2017.[68] 3GPP TS 38.211, “Physical channels and modulation,”
V15.2.0 , July2018.[69] Z. Zhang, X. Chai, K. Long, A. V. Vasilakos, and L. Hanzo, “Fullduplex techniques for 5G networks: self-interference cancellation,protocol design, and relay selection,”
IEEE Communications Magazine ,vol. 53, no. 5, pp. 128–137, May 2015.[70] Z. Ni, J. A. Zhang, X. Huang, K. Yang, and F. Gao, “Parameterestimation and signal optimization for joint communication and radarsensing,” in , June 2020, pp. 1–6.[71] L. Zheng and X. Wang, “Super-resolution delay-doppler estimationfor OFDM passive radar,”
IEEE Transactions on Signal Processing ,vol. 65, no. 9, pp. 2197–2210, May 2017.[72] H. Chu, L. Zheng, and X. Wang, “Super-resolution mmwave channelestimation for generalized spatial modulation systems,”
IEEE Journalof Selected Topics in Signal Processing , vol. 13, no. 6, pp. 1336–1347,2019.[73] J. Deng, O. Tirkkonen, and C. Studer, “Mmwave channel estimation viaatomic norm minimization for multi-user hybrid precoding,” in ,April 2018, pp. 1–6.[74] Jun Li, Guisheng Liao, and H. Griffiths, “Range-dependent cluttercancellation method in bistatic MIMO-STAP radars,” in
Proceedings of2011 IEEE CIE International Conference on Radar , vol. 1, Oct 2011,pp. 59–62.[75] L. Qian, “Radar clutter suppression solution based on ICA,” in , Nov 2013, pp. 429–432.[76] G. Marino, D. Tarchi, V. Kyovtorov, Sammartino, and P. F., “Groundbased MIMO SAR and land clutter modelling: difficulties and guide-lines,” in , 2015, Con-ference Proceedings, pp. 1–5.[77] V. F. Mecca, D. Ramakrishnan, and J. L. Krolik, “MIMO radar space-time adaptive processing for multipath clutter mitigation,” in
FourthIEEE Workshop on Sensor Array and Multichannel Processing, 2006. ,July 2006, pp. 249–253.[78] C. Liu, C. Song, and Q. Lu, “Random noise de-noising and directwave eliminating based on SVD method for ground penetrating radarsignals,”
Journal of Applied Geophysics , vol. 144, pp. 125 – 133, 2017.[79] Y. C. Eldar, R. Levi, and A. Cohen, “Clutter removal in sub-nyquistradar,”
IEEE Signal Processing Letters , vol. 22, no. 2, pp. 177–181,Feb 2015.[80] P. K. Verma, A. N. Gaikwad, D. Singh, Nigam, and M. J., “Analysis ofclutter reduction techniques for through wall imaging in UWB range,”
Progress In Electromagnetics Research B , vol. 17, pp. 29–48, 2009.[81] F. Tivive, A. Bouzerdoum, and M. Amin, “A subspace projectionapproach for wall clutter mitigation in through-the-wall radar imaging,”
IEEE Trans. Geosci. Remote Sens. , vol. 53, no. 4, p. 21082122, 2015.[82] A. M. Haimovich, R. S. Blum, and L. J. Cimini, “MIMO radar withwidely separated antennas,”
IEEE Signal Processing Magazine , vol. 25,no. 1, pp. 116–129, 2008. [83] A. Hassanien and S. A. Vorobyov, “Phased-mimo radar: A tradeoffbetween phased-array and mimo radars,”
IEEE Transactions on SignalProcessing , vol. 58, no. 6, pp. 3137–3151, June 2010.[84] N. Gonzlez-Prelcic, R. Mndez-Rial, and R. W. Heath, “Radar aidedbeam alignment in mmwave v2i communications supporting antennadiversity,” in , 2016, pp. 1–7.[85] A. Khawar, “Spectrum Sharing between Radar and CommunicationSystems,”
PhD thesis , 2015.[86] M. Bica, K. Huang, V. Koivunen, and U. Mitra, “Mutual informationbased radar waveform design for joint radar and cellular communica-tion systems,” in , March 2016, pp. 3671–3675.[87] M. R. Bell, “Information theory and radar waveform design,”
IEEETrans. Inf. Theory , vol. 39, no. 5, pp. 1578–1597, Sept. 1993.[88] Y. Yang and R. S. Blum, “MIMO radar waveform design based onmutual information and minimum mean-square error estimation,”
IEEETrans. Aerosp. Electron. Syst. , vol. 43, no. 1, pp. 330–343, Jan. 2007.[89] Z. Zhu, S. Kay, and R. S. Raghavan, “Information-theoretic optimalradar waveform design,”
IEEE Signal Process. Lett. , vol. 24, no. 3, pp.274–278, Mar. 2017.[90] T. Tian, T. Zhang, L. Kong, G. Cui, and Y. Wang, “Mutual informationbased partial band coexistence for joint radar and communicationsystem,” in , April 2019,pp. 1–5.[91] R. Xu, L. Peng, W. Zhao, and Z. Mi, “Radar mutual information andcommunication channel capacity of integrated radar-communicationsystem using MIMO,”
ICT Express
Signal Processing
IEEE Trans. SignalProcess. , vol. 58, no. 9, pp. 4684–4697, Sept. 2010.[95] Y. Liu, G. Liao, Z. Yang, and J. Xu, “Multiobjective optimal waveformdesign for ofdm integrated radar and communication systems,”
SignalProcessing
IEEE WirelessCommunications Letters , vol. 9, no. 2, pp. 194–197, 2020.[97] J. A. Zhang, X. Huang, Y. J. Guo, J. Yuan, and R. W. Heath Jr,“Multibeam for joint communication and sensing using steerable ana-log antenna arrays,” arXiv preprint, https://arxiv.org/abs/1810.04105.Accepted for publication in IEEE Trans. Veh. Technol.. , 2018.[98] J. A. Zhang, X. Huang, and V. Dyadyuk, “Massive hybrid antennaarray for millimeter-wave cellular communications,”
IEEE WirelessCommunications , vol. 22, no. 1, pp. 79–87, 2015.[99] C. Zhou, Y. Gu, X. Fan, Z. Shi, G. Mao, and Y. D. Zhang, “Direction-of-arrival estimation for coprime array via virtual array interpolation,”
IEEE Transactions on Signal Processing , vol. 66, no. 22, pp. 5956–5971, 2018.[100] X. Wang, A. Hassanien, and M. G. Amin, “Sparse transmit array designfor dual-function radar communications by antenna selection,”
DigitalSignal Process. , vol. 83, pp. 223–234, 2018.[101] D. Ma, N. Shlezinger, T. Huang, Y. Shavit, M. Namer, Y. Liu, and Y. C.Eldar, “Spatial modulation for joint radar-communications systems:Design, analysis, and hardware prototype,” 2020.[102] A. R. Chiriyath and D. W. Bliss, “Effect of clutter on joint radar-communications system performance inner bounds,” in , Nov 2015,pp. 1379–1383.[103] A. R. Chiriyath, B. Paul, and D. W. Bliss, “Joint radar-communicationsinformation bounds with clutter: The phase noise menace,” in , May 2016, pp. 1–6.[104] A. R. Chiriyath, B. Paul, and D. W. Bliss, “Simultaneous radardetection and communications performance with clutter mitigation,”in , May 2017, pp. 0279–0284. [105] S.-W. Chen, L. K. Wang, and J.-H. Lan, “Moving Object trackingBased on Background Subtraction Combined Temporal Difference,” in International Conference on Emerging Trends in Computer and ImageProcessing (ICETCIP’2011) , 2011, Conference Proceedings, pp. 16–19.[106] T. Wei and X. Zhang, “mTrack: High-Precision Passive Tracking UsingMillimeter Wave Radios,” in
MobiCom15 . ACM, 2015, ConferenceProceedings, pp. 117–129.[107] A. Sobral and A. Vacavant, “A comprehensive review of backgroundsubtraction algorithms evaluated with synthetic and real videos,”
Com-puter Vision and Image Understanding , vol. 122, pp. 4 – 21, 2014.[108] Y. Zhao and Y. Su, “Vehicles detection in complex urban scenes usinggaussian mixture model with fmcw radar,”
IEEE Sensors Journal ,vol. 17, no. 18, pp. 5948–5953, 2017.[109] M. Alfakih, M. Keche, and H. Benoudnine, “Gaussian mixture mod-eling for indoor positioning wifi systems,” in ,2015, pp. 1–5.[110] L. Haihan, Y. Li, S. Zhou, and W. Jing, “A novel method to obtain CSIbased on gaussian mixture model and expectation maximization,” in , Oct 2016, pp. 1–5.[111] X. Wu, W. Zhu, and J. Yan, “A fast gridless covariance matrix re-construction method for one- and two-dimensional direction-of-arrivalestimation,”
IEEE Sensors Journal , vol. 17, no. 15, pp. 4916–4927,2017.[112] M. Srensen and L. De Lathauwer, “Multiple invariance esprit fornonuniform linear arrays: A coupled canonical polyadic decompositionapproach,”
IEEE Transactions on Signal Processing , vol. 64, no. 14,pp. 3693–3704, 2016.[113] Y. Wang, Y. Zhang, Z. Tian, G. Leus, and G. Zhang, “Super-resolutionchannel estimation for arbitrary arrays in hybrid millimeter-wavemassive mimo systems,”
IEEE Journal of Selected Topics in SignalProcessing , vol. 13, no. 5, pp. 947–960, 2019.[114] M. A. Davenport., M. F. Duarte, Y. C. Eldar, and G. Kutyniok,
Introduction to compressed sensing . New York: Cambridge UniversityPress, 2012.[115] W. U. Bajwa, J. Haupt, A. M. Sayeed, and R. Nowak, “CompressedChannel Sensing: A New Approach to Estimating Sparse MultipathChannels,”
Proceedings of the IEEE , vol. 98, no. 6, pp. 1058–1076,2010.[116] S. Ji, Y. Xue, and L. Carin, “Bayesian Compressive Sensing,”
IEEETransactions on Signal Processing , vol. 56, no. 6, pp. 2346–2356, 2008.[117] X. Cheng, M. Wang, and S. Li, “Compressive Sensing Based Beam-forming for Millimeter-Wave OFDM Systems,”
IEEE Transactions onCommunications , vol. 65, no. 1, pp. 371 – 386, 2017.[118] Z. Zhang and B. D. Rao, “Extension of SBL algorithms for the recoveryof block sparse signals with intra-block correlation,”
IEEE Transactionson Signal Processing , vol. 61, no. 8, pp. 2009–2015, April 2013.[119] R. W. Heath, N. Gonzalez-Prelcic, S. Rangan, W. Roh, and A. M.Sayeed, “An Overview of Signal Processing Techniques for MillimeterWave MIMO Systems,”
IEEE Journal of Selected Topics in SignalProcessing , vol. 10, no. 3, pp. 436–453, 2016.[120] M. A. Hadi, S. A. Alshebeili, K. Jamil, and F. E. A. El-Samie,“Compressive sensing applied to radar systems: an overview,”
Signal,Image and Video Processing , vol. 9, pp. 25–39, 2015.[121] C. F. Caiafa and A. Cichocki, “Computing sparse representations ofmultidimensional signals using Kronecker bases,”
Neural Computation ,vol. 25, no. 1, pp. 186–220, Jan 2013.[122] Y. Sun, T. Fei, and N. Pohl, “A high-resolution framework for range-doppler frequency estimation in automotive radar systems,”
IEEESensors Journal , vol. 19, no. 23, pp. 11 346–11 358, 2019.[123] Z. Zhang, T. Jung, S. Makeig, and B. D. Rao, “Compressed sensingfor energy-efficient wireless telemonitoring of noninvasive fetal ECGvia block sparse bayesian learning,”
IEEE Transactions on BiomedicalEngineering , vol. 60, no. 2, pp. 300–309, Feb 2013.[124] Y. Li, X. Wang, and Z. Ding, “Multi-target position and velocityestimation using ofdm communication signals,”
IEEE Transactions onCommunications , vol. 68, no. 2, pp. 1160–1174, 2020.[125] Y. Chi, L. L. Scharf, A. Pezeshki, and A. R. Calderbank, “Sensitivity tobasis mismatch in compressed sensing,”
IEEE Transactions on SignalProcessing , vol. 59, no. 5, pp. 2182–2195, May 2011.[126] G. Tang, B. N. Bhaskar, P. Shah, and B. Recht, “Compressed SensingOff the Grid,”
IEEE Transactions on Information Theory , vol. 59,no. 11, pp. 7465–7490, 2013.[127] P. Zhang, L. Gan, S. Sun, Ling, and C., “Atomic norm denoising-based channel estimation for massive multiuser MIMO systems,” in , 2015,Conference Proceedings, pp. 4564–4569.[128] W. Cui, Q. Shen, W. Liu, and S. Wu, “Low complexity doa estimationfor wideband off-grid sources based on re-focused compressive sensingwith dynamic dictionary,”
IEEE Journal of Selected Topics in SignalProcessing , vol. 13, no. 5, pp. 918–930, 2019.[129] A. Fannjiang and H.-C. Tseng, “Compressive radar with off-gridtargets: a perturbation approach,”
Inverse Problems , vol. 29, no. 5, p.054008, 2013. [Online]. Available: http://stacks.iop.org/0266-5611/29/i=5/a=054008[130] X. Cheng, C. Tang, and Z. Zhang, “Accurate channel estimationfor millimeter-wave mimo systems,”
IEEE Transactions on VehicularTechnology , vol. 68, no. 5, pp. 5159–5163, 2019.[131] Y. Chi and M. Ferreira Da Costa, “Harnessing sparsity over thecontinuum: Atomic norm minimization for superresolution,”
IEEESignal Processing Magazine , vol. 37, no. 2, pp. 39–57, 2020.[132] W. Tang, H. Jiang, and S. Pang, “Grid-free dod and doa estimation formimo radar via duality-based 2d atomic norm minimization,”
IEEEAccess , vol. 7, pp. 60 827–60 836, 2019.[133] Y. Tsai, L. Zheng, and X. Wang, “Millimeter-wave beamformed full-dimensional mimo channel estimation based on atomic norm mini-mization,”
IEEE Transactions on Communications , vol. 66, no. 12, pp.6150–6163, 2018.[134] D. Shutin, “Cluster analysis of wireless channel impulse responses,”in
International Zurich Seminar on Communications, 2004 , Feb 2004,pp. 124–127.[135] C. W. Lim and M. B. Wakin, “Recovery of periodic clustered sparsesignals from compressive measurements,” in , Dec 2014,pp. 409–413.[136] L. Yu, H. Sun, G. Zheng, and J. Pierre Barbot, “Model basedbayesian compressive sensing via local beta process,”
Signal Process. ,vol. 108, no. C, pp. 259–271, Mar. 2015. [Online]. Available:http://dx.doi.org/10.1016/j.sigpro.2014.09.018[137] L. Yu, C. Wei, J. Jia, and H. Sun, “Compressive sensing for clusterstructured sparse signals: variational bayes approach,”
IET SignalProcessing , vol. 10, no. 7, pp. 770–779, 2016.[138] Z. Zhang and B. D. Rao, “Recovery of block sparse signals usingthe framework of block sparse bayesian learning,” in , March 2012, pp. 3345–3348.[139] C. R. Berger, B. Demissie, J. Heckenbach, P. Willett, and S. Zhou,“Signal processing for passive radar using OFDM waveforms,”
IEEEJournal of Selected Topics in Signal Processing , vol. 4, no. 1, pp. 226–238, Feb. 2010.[140] X. Li, D. Zhang, Q. Lv, J. Xiong, S. Li, Y. Zhang, and H. Mei,“Indotrack: Device-free indoor human tracking with commodity Wi-Fi,”
Proc. ACM Interact. Mob. Wearable Ubiquitous Technol. , vol. 1,no. 3, Sep. 2017. [Online]. Available: https://doi.org/10.1145/3130940[141] K. Qian, C. Wu, Y. Zhang, G. Zhang, Z. Yang, and Y. Liu, “Widar2.0:Passive human tracking with a single Wi-Fi link,” in
Proceedingsof the 16th Annual International Conference on Mobile Systems,Applications, and Services . New York, NY, USA: Associationfor Computing Machinery, 2018, p. 350361. [Online]. Available:https://doi.org/10.1145/3210240.3210314[142] H. Huang, Y. Peng, J. Yang, W. Xia, and G. Gui, “Fast beamformingdesign via deep learning,”
IEEE Transactions on Vehicular Technology ,vol. 69, no. 1, pp. 1065–1069, Jan 2020.[143] J. Sun, W. Shi, Z. Yang, J. Yang, and G. Gui, “Behavioral modelingand linearization of wideband rf power amplifiers using bilstm networksfor 5g wireless systems,”
IEEE Transactions on Vehicular Technology ,vol. 68, no. 11, pp. 10 348–10 356, Nov 2019.[144] F. Tang, Y. Kawamoto, N. Kato, and J. Liu, “Future intelligent andsecure vehicular network toward 6g: Machine-learning approaches,”
Proceedings of the IEEE , vol. 108, no. 2, pp. 292–307, Feb 2020.[145] B. Mao, F. Tang, Z. M. Fadlullah, and N. Kato, “An intelligentroute computation approach based on real-time deep learning strategyfor software defined communication systems,”
IEEE Transactions onEmerging Topics in Computing , pp. 1–1, 2019.[146] N. Kato, Z. M. Fadlullah, B. Mao, F. Tang, O. Akashi, T. Inoue, andK. Mizutani, “The deep learning vision for heterogeneous networktraffic control: Proposal, challenges, and future perspective,”
IEEEWireless Communications , vol. 24, no. 3, pp. 146–153, June 2017.[147] W. Wang, A. X. Liu, M. Shahzad, K. Ling, and S. Lu, “Device-free human activity recognition using commercial wifi devices,”
IEEEJournal on Selected Areas in Communications , vol. 35, no. 5, pp. 1118–1131, May 2017. [148] G. Wang, Y. Zou, Z. Zhou, K. Wu, and L. M. Ni, “We can hear youwith wi-fi!” IEEE Transactions on Mobile Computing , vol. 15, no. 11,pp. 2907–2920, Nov 2016.[149] Y. Wang, K. Wu, and L. M. Ni, “Wifall: Device-free fall detection bywireless networks,”
IEEE Transactions on Mobile Computing , vol. 16,no. 2, pp. 581–594, Feb 2017.[150] S. Yousefi, H. Narui, S. Dayal, S. Ermon, and S. Valaee, “A surveyon behavior recognition using WiFi channel state information,”
IEEECommunications Magazine , vol. 55, no. 10, pp. 98–104, Oct 2017.[151] Z. Shi, J. A. Zhang, R. Xu, and Q. Cheng, “Deep learning networks forhuman activity recognition with csi correlation feature extraction,” in
ICC 2019 - 2019 IEEE International Conference on Communications(ICC) , May 2019, pp. 1–6.[152] A. Al-Hourani, R. J. Evans, S. Kandeepan, B. Moran, and H. El-tom, “Stochastic geometry methods for modeling automotive radarinterference,”
IEEE Transactions on Intelligent Transportation Systems ,vol. 19, no. 2, pp. 333–344, 2018.[153] P. Chu, A. Zhang, X. Wang, Z. Fei, G. Fang, and D. Wang, “Inter-ference characterization and power optimization for automotive radarwith directional antenna,”
IEEE Transactions on Vehicular Technology ,pp. 1–1, 2020.[154] J. Liang and Q. Liang, “Design and analysis of distributed radar sensornetworks,”
IEEE Transactions on Parallel and Distributed Systems ,vol. 22, no. 11, pp. 1926–1933, Nov 2011.[155] N. Gonzlez-Prelcic, R. Mndez-Rial, and R. W. Heath, “Radar aidedbeam alignment in mmWave V2I communications supporting antennadiversity,” in , Jan 2016, pp. 1–7.[156] C. Jiao, Z. Zhang, C. Zhong, and Z. Feng, “An indoor mmwave jointradar and communication system with active channel perception,” in2018 IEEE International Conference on Communications (ICC)