A Survey of Channel Modeling for UAV Communications

0 A Surv e y of Channel Modeling for U A V Communications Aziz Altaf Khuwaja , Y unfei Chen , Senior Member , IEEE, Nan Zhao , Senior Member , IEEE, Mohamed-Slim Alouini , F ellow , IEEE, Paul Dobbins Aziz Altaf Khuwaja is with the School of Engineering, University of W arwick, Coventry , U.K. CV4 7AL and also with the Department of Electrical Engineering, Sukkur IB A Univ ersity , Sukkur, Sindh, P akistan (e-mail: A.khuwaja@warwick.ac.uk) Y unfei Chen is with the School of Engineering, Univ ersity of W arwick, Cov entry , U.K. CV4 7AL (e-mail: Y un- fei.Chen@warwick.ac.uk) Nan Zhao is with the School of Information and Communication Engineering, Dalian Univ ersity of T echnology , China (e- mail: zhaonan@dlut.edu.cn) Mohamed-Slim Alouini is with the King Abdullah University of Science and T echnology (KA UST), Thuwal 23955-6900, Makkah Province, Kingdom of Saudi Arabia (e-mail: slim.alouini@kaust.edu.sa) Paul Dobbins is with the T elent, Haywood Road, W arwick, CV34 5AH, UK (e-mail: paul.dobbins@telent.com) DRAFT January 24, 2018 1 Abstract Unmanned aerial vehicles (U A Vs) ha ve gained great interest for rapid deplo yment in both ci vil and military applications. U A V communication has its o wn distinctiv e channel characteristics compared with widely used cellular and satellite systems. Thus, accurate channel characterization is crucial for the performance optimization and design of efﬁcient UA V communication systems. Howe ver , se veral challenges exist in U A V channel modeling. For example, propagation characteristics of U A V channels are still less explored for spatial and temporal v ariations in non–stationary channels. Also, airframe shadowing has not yet been in vestigated for small size rotary UA Vs. This paper provides an extensi ve surve y on the measurement campaigns launched for U A V channel modeling using low altitude platforms and discusses v arious channel characterization ef forts. W e also re view the contemporary perspecti ve of U A V channel modeling approaches and outline some future research challenges in this domain. Index T erms Channel characterization, channel models, measurement campaigns, U A V communication. I . I N T R O D U C T I O N Unmanned aerial vehicles (U A V) aided communication has seen drastic de velopment in a v ariety of applications. F or instance, it can be used in military operations for border surveil- lance and information gathering in hostile en vironment. Also, it can be deployed to monitor ci vil emer gency services, conduct humanitarian missions and facilitate scientiﬁc data collection. Most of these applications deploy U A Vs as low altitude platforms, and they can be fully au- tonomous or remotely operated. In order to ensure safety and high reliability , it is of utmost importance to thoroughly characterize the U A V channels particularly for lo w altitude platforms. Many research organizations and standardization bodies ha ve w orked together to establish prag- matic U A V frame works. For example, special committee (SC–228) has been formed by the Radio T echnical Commission for Aeronautics (R TCA) in 2013 to frame minimum performance standards for U A V operations [1]. R TCA has also established drone advisory committee in 2016 to ensure safe introduction of U A Vs into the national airspace system [2]. Also, National Aero- nautics and Space Administration (NASA) and Federal A viation Administration (F AA) hav e launched a joint research initiati ve to integrate U A Vs in national airspace system across the United States [3]. The most unique features that distinguish U A V communication systems from con ventional wire- January 24, 2018 DRAFT 2 less communication systems or the unique channel characteristics of U A V communications are: • the distinct communication channels, i.e., air–ground (AG) and air–air (AA) channels • the spatial and temporal variations in non–stationary channels • the airframe shadowing caused by the structural design and rotating capability of U A V . In a di verse propagation en vironment where the U A V operates, these features become more challenging. Propagation characteristics for terrestrial cellular systems are often corroborated with well–established empirical and analytical models. The satellite links for land mobile sys- tems ha ve also been thoroughly in vestigated in the literature [4], [5]. Ho we ver , these models are often not well suited to characterize the UA V propagation channel due to the aforementioned unique attributes of theirs. On the other hand, reliable analytical models are necessary to e valuate the performance of dif- ferent wireless techniques and to provide assistance to link b udget calculations. In the context of A G channels in U A V communications, modeling approaches can be generally classiﬁed in three categories. The ﬁrst approach is to dev elop deterministic models using en vironmental parame- ters, while considering the U A V altitude and ele v ation angle from the ground. Such models are useful to study the fading effects in the channel [6], [7], the propagation conditions [8], [9] and hence can provide co verage analysis for optimal U A V position [10], [11]. The second approach is to dev elop the tapped delay line (TDL) model to characterize the direct path as well as the multipath components. This giv es the wideband frequency–selecti ve parameters deri ved from channel impulse response [12]– [14]. This approach is particularly important if non–stationarity in A G channel is to be addressed. Finally , the geometric–based stochastic model is desirable for ev aluating spatial–temporal characteristics in a geometric simulation en vironment. This ap- proach is more preferable to characterize the A G channel in 3D plane with less en vironmental parameters [15]– [19]. Ho wev er , analytical models alone do not always describe the real beha vior of the propagation channel, because of deﬁcient realistic assumptions. Therefore, empirical studies initiated by measurement campaigns are essential. Most of the works reported in the literature [20]– [34] are pertinent to the A G channel characterization based on the measurement campaigns launched with manned aircrafts in high altitude platform. Howe ver , these ﬁndings cannot be directly DRAFT January 24, 2018 3 applied to the single–hop U A V network deployed in lo w altitude platform, i.e., up to 120 m as permitted by F AA in USA [35] and Ci vil A viation Safety Authority (CASA) in Australia [36]. It is e vident from the studies in [37]– [57] that the impact of airborne platforms are signiﬁcant for the channel characteristics of U A V communications. Moreover , less research ef forts ha ve been made to tackle with the shadowing induced in A G and AA channels by the U A V structural de- sign and maneuv ering. In addition, the wide sense stationary uncorrelated scattering (WSSUS) assumption may be violated in some UA V–aided applications. Thus, in order to av oid exag- gerated performance ev aluation from the analytical channel model, it is important to estimate the fading statistics with stationary interv als. The AA channel in multi–hop U A V network has been empirically characterized in the literature with lo w po wer radios based on IEEE 802.15.4 [46]– [48] and IEEE 802.11 standards [49]– [50]. But these studies only reported large–scale fading statistics and the impact of antenna orientation on channel performance, thus the Doppler spectrum for AA channel is understudied. Despite the importance of channel modeling in U A V communications, v ery few surve y stud- ies are a vailable in the literature. F or instance, reference [58] identiﬁed key issues related to the formation of multi–U A V network, but this survey focuses more on the communications and especially the control of U A V . Also, aerial networking characteristics and requirements are re- vie wed in [59] for civil applications. Howe ver , this surve y mainly discussed the communications aspects of U A V , in particularly network layer designs. Both [58] and [59] ha ve very light touch on the channel modeling. On the other hand, the physical layer characterization of the A G chan- nel at L and C bands w as comprehensi vely re viewed in [60]. Howe ver , measurement campaigns in this paper were reported mainly for aeronautical communications and land mobile satellite systems at the L and C bands. In contrast, our survey will re view the current advances for the U A V channel characterization. The rest of the paper is or ganized as follows. In Section II, we will revie w the measurement campaigns launched with U A Vs as low altitude platforms, where we will categorize them ac- cording to the bandwidth of channel sounder , the lo w–power and low–cost radios and the widely deployed cellular infrastructure. Characterization of AA and A G propagation using empirical channel models will be discussed in Section III. In Section IV , we will categorize analytical U A V channel models as deterministic, stochastic and geometry–based stochastic. In Section January 24, 2018 DRAFT 4 V , we will highlight some important issues pertinent to the airframe shadowing, non–stationary channels and the applicability of di versity techniques in U A V communications. Finally , we will discuss future research challenges in U A V channel modeling. I I . M E A S U R E M E N T C A M P A I G N S Propagation channel models dev eloped using analytical approaches do not always give sat- isfactory performances in real–time deployment due to inadequate or unrealistic assumptions. In this case, the actual behavior of the propagation channel can only be understood via ﬁeld measurements. A number of measurement campaigns have been carried out in di verse en viron- ments to understand the U A V channel characterization. Some of these campaigns analyze the A G propagation channel with lar ge–scale and small–scale fading effects and hence only propose empirical channel models. Other measurement campaigns assess the use of div ersity techniques for range extension and enhancement of communication channel throughputs. In the literature, most of the measurement campaigns hav e been conducted using two types of aerial vehicles. The ﬁrst type of aerial vehicles are small and medium sized manned aircrafts. For instance, in [12]– [14], S–3B V iking aircraft was used to comprehend the A G channel char - acteristics at L and C bands in different en vironments. In [25], the Cessna-172S aircraft was used to e valuate the performance of a 4 × 4 MIMO enabled OFDM system for the A G chan- nel. In [26] and [27], UH–1H military helicopter was used to study the A G channel in a 4 × 2 MIMO conﬁguration to achiev e the div ersity gain and equalization ef ﬁcacy to mitigate ISI in frequency–selecti ve channels. In [28], a news–reporting helicopter w as used to attain spatial multiplexing gain and throughputs for airborne communication in 2 × 2 MIMO settings. The lo- gistics in volv ed in the measurement campaigns using manned aircrafts are expensi ve and daunt- ing. Therefore, the second type of aerial vehicles i.e., U A Vs are preferable to reduce the cost. In this case, the U A V payload is often integrated with an on–board processor to control ﬂight dynamics and wireless equipment to collect measurement data. In addition, the experimental setup also contains antennas to radiate and receiv e RF signals, GPS system to record telemetry data and inertial measurement unit (IMU) to measure ﬂight dynamics such as pitch, ya w and roll angles. In the rest, we will mainly focus on the measurement campaigns using U A Vs and will categorize them into three main groups based on the wireless techniques. The ﬁrst group of the DRAFT January 24, 2018 5 measurement campaigns use narro wband or wideband channel sounder . The second set of mea- surement campaigns use IEEE 802.11 radios. In the third group, the measurement campaigns are accomplished using the widely deployed cellular network infrastructure. A. Narr owband and W ideband Channel Sounder 1) Narr owband Measur ement Systems: They ev aluate the Doppler frequency shift and the channel gain experienced by a narrowband continuous wav e (CW) signals. These systems have a channel sounder that generates pilot tones at a single carrier frequency using a CW generator . Examples of narrowband measurement campaigns for characterizing the A G propagation chan- nels in aeronautical communications for very high frequency (VHF) band are [29], [30], for L band [31] and for higher frequency (HF) band [32]. In [37], the measurement campaign was performed in an urban area of Prague, Czech Repub- lic, using 2 GHz CW transmitter with a bandwidth of 12.5 kHz. The measurement setup includes an airship U A V mounted with transmitter and four–channel custom–made recei ver located at the ground station. Also, monopole antennas were used at both the transmitter and receiv er sides. The UA V ﬂe w between 100 to 170 m above the ground lev el at lo w ele vation angle between 1 ◦ to 6 ◦ . The authors hav e statistically characterized the A G channel which ﬁts in between a purely terrestrial link and a land mobile satellite system. They hav e also presented a narrowband channel estimator capable of replicating signal dynamics. Some related measurement campaigns were conducted with similar equipment in Prague for path loss model in urban area [38] with a ﬂight altitude between 150 to 300 m. Further , measurements in [61] and [62] were conducted in urban and wooded areas, respecti vely , to study space di versity techniques for similar situations. In [39], ﬁeld e xperiments were performed in a sub urban terrain of Madrid, Spain, with two fre- quency bands i.e., 5.76 GHz for narrowband measurements. Field trials were carried out using a hexacopter U A V , univ ersal software deﬁned radio peripheral (USRP) hardware, and clover - leaf antennas with circular polarization. In the narrowband measurement scheme, frequency modulated CW transmitter was used for both vertical and horizontal ﬂight routes in order to characterize the lar ge–scale variation. In this case, USRP was used as the ground receiver . The U A V ﬂew at an altitude between 0 to 50 m for the vertical ﬂight route and co vered distance of 20 m and 30 m for horizontal route. The authors hav e in vestigated large–scale fading ef fects in U A V propagation channel and computed path loss exponent for both vertical and horizontal January 24, 2018 DRAFT 6 ﬂight direction using the dual slope and the log–distance path loss models, respectively . They found that for the vertical ﬂight direction, the attenuation decreased below the breakpoint dis- tance and then increased with U A V altitude. Whereas, the attenuation increased exponentially with the horizontal ﬂight direction. Furthermore, the y hav e also modeled f ast f ading effects with Rician distribution. Narro wband measurement systems are appropriate for computing frequency non–selecti ve fading parameters because of the limited channel sounding spectrum. Howe ver , these systems may not be preferred in rich multipath en vironment to characterize the performance of U A V propagation for coherence bandwidth and multipath delay statistics. Also, the performance ev al- uation of MIMO capacity using these systems may seems to be a difﬁcult task as they are not able to resolve indi vidual multipath components. 2) W ideband Measurement Systems: The y determine the channel impulse response (or trans- fer function) and frequency–selecti ve parameters deri ved from it. Wideband channel measure- ments for characterizing the aeronautical propagation channels are mostly conducted with spread spectrum channel sounder . One such type is correlativ e channel sounder , where pseudonoise (PN) sequence is transmitted as the channel sounding signal and the receiv ed signal is then cor- related at the receiver with the same PN sequence. For example, correlativ e channel sounder was used in [33] and [34] for measuring multipath ef fects. In the conte xt of characterizing the U A V propagation channel, wideband frequency–selecti ve parameters are often measured with USRP platforms, for instance, as used in [39] and [63]. This platform provides more ﬂexibility in terms of lo w–power consumption and multiple frequenc y bands. In [39], the wideband measurement campaign was also performed with the channel sounding signal generated by the L TE base station at the frequency of 1.817 GHz. In this case, USRP w as mounted on the U A V as the recei ver module with antennas placed below the U A V propellers. In this work, the small–scale v ariations in the U A V propagation channel was characterized with the measured channel impulse responses and estimated delay spread and power delay proﬁle. The authors hav e analyzed channel statistics using the cumulativ e distribution function (CDF) and observed the random beha vior of the multipath components at dif ferent U A V altitude. In [40], the measurement campaign was conducted for open and sub urban spaces on the cam- pus of Florida International University . The experimental setup consisted of a quadcopter U A V , the ultra–wideband (UWB) channel sounding radio tuned at the frequency of 4.3 GHz with a DRAFT January 24, 2018 7 range between 3.1 GHz to 5.3 GHz and the planar elliptical dipole antennas. In this setting, U A V altitude raised from 4 m to 16 m with the step size of 4 m. The same UWB radio was used as the ele v ated ground receiv er at two different heights and positioned in three different scenarios. In the ﬁrst scenario, the recei ver was placed under the tree canop y and ele vated at 1.5 m from the ground. In the second scenario, the receiv er w as placed at the same height with clear line–of–sight (LOS) to the transmitter . In the third scenario, the receiv er was laid down at 7 cm from the ground in LOS condition. In this work, the authors hav e characterized A G propagation channel, where they proposed the empirical path loss model for both static and mobile U A Vs. They found the worst path loss attenuation for mobile U A V motion in the ﬁrst scenario, whereas, the best for static U A V in the second scenario. Also, they characterized the fading channel as Nakagami m and presented multipath propagation model. In [63], the measurement campaign was performed in a residential area and mountainous desert landscape in Arizona, USA, with SDR platform tuned at 5.8 GHz. The USRP radio was attached with a octocopter U A V and also serv ed as ground base station, where both radios were equipped with the dual band vertically oriented omnidirectional antenna and controlled by open source GNU radio. The authors ha ve characterized the frequenc y–selecti vity of the A G propagation by the av erage and RMS delay spread of the channel. Also, Doppler power spectrum was calculated by summing the entire range of scattering function delay . They analyzed channel statistics with the CDF and found that the desert terrain causes substantial delay spread in the A G propagation than in a residential area. Moreov er , CDF analysis followed a log–normal trend for the RMS Doppler spread. W ideband measurement campaigns are desirable to ev aluate both narro wband and wideband frequency–selecti ve channel parameters. Ho wev er , additional computational capabilities are required to process the raw data collected from the measurements. Therefore, this type of mea- surement systems may not be suitable for real–time characterizing of fading channel parameters. Also, cost and physical dimensions of wideband channel sounding equipment are other possible constraints that need to be considered. B. IEEE 802.11 based U A V Measur ements U A V channel characterization using commercial off–the–shelf 802.11 radios are desirable due to their lo w po wer consumption, cost effecti veness and ﬂexibility to be integrated with small January 24, 2018 DRAFT 8 size U A Vs. Ho we ver , the performances of such radios are prone to interference and background noise. Also, ﬁxed narro wband frequenc y and limited communication range are other constraints to ev aluate fading channel parameters. Channel characterization ef forts reported in the literature for multi–hop U A V networks were based on IEEE 802.11 in [41]– [45] and also IEEE 802.15.4 ZigBee devices in [46]– [48]. In this section we revie w the measurement campaigns relev ant to 802.11 radios for single–hop U A V network only . In [49], the measurement campaign was performed in laboratory and outdoor en vironment to particularly study the altitude–dependent multipath propagation in AA channel. The mea- surements were collected with 802.11 a/b/g/n WLAN devices from two different vendors and deployed in three outdoor scenarios using a hexacopter U A V . The laboratory experiments were conducted for sensitivity analysis and calibration purpose. In the ﬁrst scenario, the impact of ﬂight distance followed a free space path loss model. In the second scenario, good signal re- ception was attained between 170 ◦ –230 ◦ and the worst signal for a yaw angle of 240 ◦ –260 ◦ . Finally the ef fect of the ground reﬂected multipath components on U A V altitude was examined for the ﬂight altitude between 10 to 40 m and proposed the height dependent Rician model with K factor reliant on the U A V altitude. In [50], the measurement campaign was performed in an open space using a quadcopter U A V and an access point (AP) connected with 802.11a WLAN interface. Also, IMU module was used to measure the UA V position and orientation in 3D plane. Measurements were collected with three horizontally aligned dipole antennas and at the ﬂight altitude between 15 to 110 m. The authors observed that for both A G and AA channels, the path loss exponent computed by log–distance model roughly matched with that of free space propagation. Also Nakagami m distribution was found to be a good ﬁt for a multipath fading channel. Furthermore, inter–arri val time of packet and retransmission attempts were analyzed by empirical CDFs. Related ﬁeld trials were conducted in [51] for an open space and a campus en vironment for the U A V ﬂight altitude v aries between 20 to 120 m in dif ferent testing scenarios. T wo v ertically polarized omnidirectional antennas were mounted on both U A V and AP . The authors found that the optimal antenna orientation can alle viate the impact of U A V altitude on recei ved signal strength and throughputs. Moreo ver , horizontally aligned antennas reduced the affect of U A V yaw angle on throughputs. They have also found that the propagation condition follo wed that in DRAFT January 24, 2018 9 the free space for an open ﬁeld. In [52], the measurement campaign was launched at a pri vate airﬁeld in Connecticut, USA, using a 802.11a radio mounted on a ﬁxed–wing U A V . Commercially av ailable dual–band om- nidirectional antenna and custom–made antennas tuned at 5.28 GHz were tested with 32 ori- entation pair conﬁgurations. The UA V ﬂew approximately at 64 km/h airspeed and maintained an altitude of 46 m ov er the ground receiv er nodes. The authors ev aluated the throughput re- liance with U A V transmit antenna and reported highest rates with horizontal dipole, orthogonal to ﬂight direction and parallel to the ground. In addition, the y also estimated that path loss roughly follo wed free space propagation. In [53], a related measurement campaign was performed with both 2.4 GHz 802.11g and 5.8 GHz 802.11a devices. In this case, the authors computed the maximum range attained with 802.11a radio and compared that by 802.11g. They found that 802.11g node can provide rob ust communication at the altitude of approximately 183 m. In this work, another experimental trial was conducted with 900 MHz 802.11 radio to determine recei ved signal strength and throughput performances. They found signiﬁcant communication range up to 2000 m with throughputs in Mbps. In addition, the y hav e analyzed the path loss attenuation with linear re gression method. In [64], the measurement campaign was done in the farmland area amid by woods. In this work, A G channel characterization was performed in terms of network le vel div ersity gain and found signiﬁcant enhancement in packet transmission rate by multiple recei vers. Lo w–power and cost–ef fectiv e IEEE 802.11 radios are preferable for narro wband ﬁeld mea- surements in U A V networking. Also this platform provides an opportunity for characterizing the U A V propagation channel with v arious antenna orientations. Thus, it provides the optimal placement and alignment of on–board U A V antennas. Ho we ver , in a complex communications en vironment where U A V operates, radio interference from other 802.11 equipment can be chal- lenging. In this case, one possible solution is to maintain high signal–to–noise–plus–interference ratio (SINR) at the physical layer for each aerial link in the presence of possible interference from the adjacent radio de vice. C. Cellular–Connected U A V Measur ements Cellular networks can be considered as a prospecti ve candidate to facilitate U A V applications in ci vil and commercial domains. W idely deployed cellular infrastructure can be utilized to January 24, 2018 DRAFT 10 provide reliable A G channels and hence, cut the cost of in vesting additional ground infrastruc- ture and spectrum allocation. Howe ver , since cellular–connected U A Vs depend on the cellular network and cellular infrastructure can collapse due to natural disaster , a viable fail–safe mech- anism is needed. Other challenges, such as down–tilted base station antennas, neighboring cell interference, handov er performance, multiple access, U A V mobility and link security , also need to be addressed thoroughly before the widespread implementation of U A V netw ork connected to the cellular networks. This has motiv ated sev eral mobile operators, telecommunication vendors and research org anizations to further scrutinize the propagation channel characteristics between cellular base station and airborne U A V . For example, Qualcomm T echnologies has launched ﬁeld measurements in San Die go, California, to assess the L TE network performance in lo w altitude platform using quadcopter U A V [65]. In another example [66], Ericsson and China Mobile ha ve conducted measurement trials in China’ s Jiangsu province to de velop 5G prototype enabled by drone U A Vs. In [54], the measurement campaign was launched in urban and rural scenarios in Germany to characterize the propagation channel between U A V and cellular base station, using 900 MHz GSM network and 1.9–2.2 GHz UMTS services. Field measurements were carried out with the ﬁxed–wing U A V and captiv e balloon at the altitude up to 500 m. This work ev aluated the ov erall aerial performance in terms of recei ved signal strength and handover analysis in both urban and rural scenario. The authors have made assumptions that the attenuation was independent from frequency and distance. It was found that due to signal de gradation at higher U A V altitude the av ailability of base station decreases. T o conclude, a good RF coverage was achiev ed in a rural en vironment due to less ground obstacles than in urban terrain. In [55] and [67], measurement campaigns were launched under the SAAS project (remote piloted semi–autonomous aerial surveillance system using terrestrial wireless networks) in an urban en vironment of Lisbon, Portugal to in vestigate the applicability of terrestrial cellular net- works in U A V communication. In [55], the ﬁeld trials were performed at GSM, UMTS and L TE cellular bands using spectrum analyzer and antenna tied with meteorological balloon, deployed as UA V platform. Recei ved signal po wer was recorded from the roof mounted base station at the U A V altitude between 11 to 18 m. In this work, empirical model was obtained for path loss attenuation in outdoor urban scenario. The w orst case scenario w as reported due to the radiation DRAFT January 24, 2018 11 pattern of the base station antenna where receiv ed signal po wer dropped as U A V climbed abov e the base station height. Howe ver , handov er analysis was not studied in this scenario. On the other hand, reference [67] presented multi–U A V network architecture based on cellular and IP networks. They hav e assessed the network le vel performance with quality of service measure- ments in terms of recei ved po wer , latency and jitters. In [56], the measurements campaign was performed in the rural en vironment at 800 MHz L TE networks with tw o dif ferent cellular service providers in Denmark. T wo ﬂight zones 7 km apart from each other were demarcated for the experimental site, surrounded by multiple base stations with a height between 20 to 50 m. The U A V ﬂew in circular track of 500 m diameter at the airspeed of 15 km/h and v aries altitude between 15 to 120 m. The authors ha ve found considerable reduction in path loss exponent and shado wing v ariation as U A V altitude increased. Therefore, their ﬁndings exhibited that the U A V propagation channel necessitates altitude dependent parameters for channel modeling. In [57], the measurement campaign was launched for suburban terrain in V ictoria, Australia, at 850 MHz L TE cellular network, using a quadcopter as U A V and Andriod based mobile phone was used for logging samples of receiv ed signal. The experimental site covered 12 k m 2 sur- rounded area with a single base station of height 30 m. U A V ﬂew at an a verage speed of 17 km/h at the altitude between 15 to 120 m. From the measurement result, they obtained an- gle–dependent parameters to characterize propagation channel between cellular base station and U A V in airborne platform. In [68], another measurement campaign was conducted for an open area and mock village in California, USA, at 909 MHz cellular band. Measurement setup consisted of a quadcopter U A V equipped with sensor package, transmitting radio with 13.9 m pneumatic mast serving as base station and ground controller . The U A V ﬂew at the altitude between 40 to 60 m and followed linear and radial ﬂight patterns. In this work, the authors hav e proposed the compositional path loss model to account two–ray ground reﬂection propagation and diffraction losses. Also, they identiﬁed low cov erage zones in cellular–connected U A V networks for be yond LOS operations and named this phenomena as “holes in the sky”. They pointed out that the primary causes for this phenomena are interference caused by tw o–ray ground reﬂection, diffraction losses incurred by the fresnel zone of propagation path and nulls in the antenna radiation pattern. Cellular networks seems to fulﬁll future requirements of U A V communication as they pro- January 24, 2018 DRAFT 12 vide extended co verage to the large area via hando ver between multiple base stations. But cellular networks are not designed to provide A G propagation abov e the base station height due to do wn–tilted sector antennas. Also, U A V applications such as search and rescue services and disaster management may be suffered due to infrastructure failure. In this case, aerial het- erogeneous network can be a promising fail–safe framework for enabling coexistence between terrestrial communication networks and satellite systems to ensure redundancy of U A V commu- nications. In this section, we have re vie wed the measurement campaigns using U A Vs as low altitude platforms. In T able I, we summarize the aforementioned measurement campaigns. I I I . E M P I R I C A L C H A N N E L M O D E L S F R O M M E A S U R E M E N T C A M PA I G N S Channel parameters can change frequently with time and space due to cruising capability of U A Vs. Therefore, channel characterization is an essential step to study the impact of fast spa- tial–temporal v ariations in the U A V channel and consequently to predict the performance of U A V communications. Thus, many measurements campaigns hav e been launched to corrobo- rate connections between channel parameters and e xperimental setups such as the ﬂight altitude, the elev ation angle, the separation distance between U A V and ground station and the operating en vironment to identify and model important factors that undermine the communication perfor- mances. Despite of all these efforts, there are no uniﬁed answers and conclusions still need to be established by means of reliable channel models. In this section we will re vie w the empirical models that characterize AA and A G propagation channels. A. Air–Air (AA) Channel Characterization The AA channel characterization is particularly essential in multi–U A V networks and aerial wireless sensor network applications, where the characteristics of the AA channel rely on the U A V altitude and relati ve velocity etc. In [46]– [48], channel characterization has been per - formed for aerial wireless sensor networks using IEEE 802.15.4 technology , where the AA channel was sho wn to hav e better conditions than the A G channel in terms of path loss exponent (PLE). In [46], empirical study has been conducted using micro aerial vehicles to characterize the impact of distance and antenna orientation on the receiv ed signal strength in the AA channel. The authors hav e performed linear regression on the samples of receiv ed signal and computed DRAFT January 24, 2018 13 T ABLE I M E A S U R E M E N T C A M P A I G N S Ref. Frequency U A V Scenario Altitude Channel Statistics [37] 2 GHz Airship Urban 100-170 m PDF , CDF , AFD, LCR, PSD, AF [38] 2 GHz Airship Urban 150-300 m PL [39] 5.76 GHz Hexacopter Suburban 0-50 m PL, SF , K, RMS, CDF 1.817 GHz [40] 4.3 GHz Quadcopter Open ﬁeld, 4-16 m PL, SF , µ , ξ , PDF , CDF , RMS Suburban BC [49] 2.4 GHz Hexacopter Laboratory , outdoor 10-40 m PL, P AS, K, PDF [50] 802.11a Quadcopter Open ﬁeld 15-110 m PL, P AS, CDF [51] 802.11a Quadcopter Open ﬁeld, campus area 20-100 m PL [52] 802.11a Fixed–wing Airﬁeld 46 m PL [53] 802.11a/g, Fixed–wing Airﬁeld, Rural 46 m, PL 900 MHz 107-274 m [54] GSM, UMTS Fixed–wing, captiv e bal- loon Urban, rural 0-500 m PL [55] GSM, UMTS, L TE W eather bal- loon Urban 11-18 m PL [56] L TE (800 MHz) Hexacopter Rural 15-100 m PL, SF [57] L TE (850 MHz) Quadcopter Suburban 15-120 m PL, SF [61], [62] 2 GHz Airship Urban, wooded 100-170 m CDF , DG, AFD, LCR [63] 5.8 GHz Octocopter Residential, mountainous – RMS, DS, CDF January 24, 2018 DRAFT 14 Ref. Frequency U A V Scenario Altitude Channel Statistics [64] 802.11 b/g Fixed–wing Farmland 75 m AF , DG [65] PCS, A WS, 700 MHz Quadcopter Mixed subur - ban 122 m PL, CDF [67] EDGE,HSP A+, L TE Hexacopter - 10-100 m R TT , J [68] 909 MHz Quadcopter Open ﬁeld, mock village 40-60 m PL, PES AF: correlation function, AFD: av erage fade duration, BC: coherence bandwidth, CDF: cumulati ve distribution function, DG: div ersity gain, DS: Doppler spread, J: jitters, K: Rician factor , LCR: level crossing rate, P AS: power azimuth spectrum, PDF: probability density function, PDP: power delay proﬁle, PES: po wer elevation spectrum, PL: path loss, RMS: RMS- delay spread, R TT : round trip time, SF: shadow f ading, µ , ξ : mean and standard deviation of Nakagami m factor PLE for AA, A G and ground–to–ground (GG) wireless channels. They found that the GG chan- nel performed poorly with PLE of 3.57. PLE for AA and A G channels were estimated to be 1.92 and 2.13, respectiv ely . Similarly , in [47], the recei ved signal strength for AA and AG channels decreases with the separation distance, but at a mar ginally reduced rate. The path loss exponent was estimated from the log–distance propagation model and PLE to be 0.93 and 1.50 for AA and A G propagation, respectiv ely . On the other hand, the authors of [48] hav e observed that the re- cei ved signal strength for A G, AA and ground–to–air (GA) propagation improv es with e xtended U A V altitude and deteriorates as U A V distance increases. They observed that the AA channel follo wed free space path loss with PLE of 2.05. Whereas, the presence of gray zones of com- munication leads to asymmetry in A G and GA channels with PLE of 2.32 and 2.51, respecti vely . Aerial link characterization has been conducted in [49] and [50] using IEEE 802.11 radio. In [49], the impact of UA V altitude on the AA propagation was in vestigated for large–scale v ari- ations and small–scale fading distrib ution. In this study , path loss was determined by the Friis equation with PLE of 2.6 and f ading channel distrib ution ﬁts with height–dependent Rician fac- tor K . In [50], log–distance model was used to analyze the path loss for vertical and horizontal distances. In this work, minimum mean square error (MMSE) method was utilized to compute the PLE of 2.03 and 2.01 for AA and A G channels, respectiv ely . DRAFT January 24, 2018 15 T ABLE II L A R G E – S C A L E FA D I N G S TA T I S T I C S F O R A A C H A N N E L Ref. PL model [46]- [48] P L ( dB ) = 10 α log 10 ( d ) , α = 1.922, [46], α = 0.93 [47], α = 2.05 [48] [49] RS S ( dB ) = P t + G U AV 1 + G U AV 2 + 10 log 10 ( λ 4 πd ) α , P t = 20 dBm, G U AV 1 = G U AV 2 = 5 dBi, α = 2.6, f c =2.4 GHz [50] P L ( dB ) = P L ( d 0 ) + 10 α log 10 ( d d 0 ) , d = p d 2 h + d 2 v , P L ( d 0 ) = 46.4 dB, α =2.03, d h ∈ { 0 , ..., 100 m } , d v = 50 m, d 0 = 1 m α : path loss exponent, RS S : received signal strength, d : separation distance, d 0 : reference distance, d h : horizontal- distance, d v : vertical distance, G U AV : U A V antenna gain The AA channel characterization highlights that the propagation conditions are generally de- termined by the vertical and horizontal distances between the multiple airborne U A Vs in LOS condition. Howe ver , signiﬁcant attenuation occurs for the characterization of aerial link be yond the LOS condition to maintain large communication range. Also it would be useful to study the consequences of the Doppler frequency shift as the multiple U A Vs cruises with higher veloci- ties. Large–scale fading statistics of the AA channel are summarized in T able II. B. Air–Gr ound (AG) Channel Characterization 1) Lar ge–Scale F ading Statistics: Most of the A G channel measurements focus on the large–scale statistics such as path loss exponent and shadow fading. For an urban en viron- ment in [38], the measured results exhibited that the path loss follows a distance–independent trend and is signiﬁcantly af fected by the lo w elev ation angle. The excess path loss model is de veloped by extending terrestrial macro cell models that includes the reﬂection and diffraction losses caused by the surrounding buildings and incorporated by the knife–edge diffraction the- ory . For a suburban en vironment in [39], the impact of U A V altitude and distance on path loss was analyzed. For U A V altitude, simpliﬁed dual slope path loss model was considered and found that PLE is neg ativ e below the breakpoint altitude because of partially cleared ﬁrst Fresnel zone, whereas when the U A V altitude increases abo ve the breakpoint le vel path loss is roughly similar to the free space propagation due to sufﬁciently cleared ﬁrst Fresnel zone. For horizontal U A V January 24, 2018 DRAFT 16 distance, path loss analyzed with log–distance model. Also, in [40], the ef fect of U A V altitude and the optimal placement of ground recei ver on path loss was stochastically modeled for both static and mobile UA V in an open ﬁeld and suburban scenario, while considering foliage losses and Doppler frequency shift. In addition, shado w fading is modeled with zero–mean Gaussian distribution and analyzed with PDF . Another empirical study was conducted in [41], to ev aluate the inﬂuence of distance on path loss attenuation and found degraded performance of the A G channel due to detrimental effect of interference from the other 802.11 de vices operated in the surrounding test area. Moreover , in [42], recei ved signal strength declined with the distance and follo wed the Friis channel model. In [43], the A G propagation channel in the single–hop U A V system follo wed log–distance model, where higher throughputs were attained over longer distance. For an open ﬁeld and campus en vironment in [51], path loss w as ev aluated with the free space path loss model. In [52] and [53], PLE w as estimated using linear regression. In [54], distance and frequenc y independent empirical path loss model was proposed for urban and rural terrains, where the altitude of aerial mobile station was accounted as the k ey modeling parameter . In contrast, the empirical propagation model in [55] suggested that path loss model is dependent on the distance in 3D plane and the operating frequency . In this case, other modeling parameters such as the U A V altitude and the tilt angle of base station sector antenna were also considered. The altitude dependent path loss model was proposed in [56], where path loss and shadow fad- ing were decreased as the U A V altitude increased from 15 to 120 m and at about 100 m the propagation condition matched to that of free space. In [57], the angle–dependent AG propaga- tion channel model w as presented, which encompasses excess path loss attenuation and shadow fading model. In this work, the model parameters are dependent on the angle between cellular base station and airborne U A V . The analytical path loss model was used in [65] to e v aluate the performance of L TE network with U A V platform, where most of the path loss samples computed by measurements were lumped between the reference PLE of 2.0 and 4.0. In [68], the combina- tional model w as de veloped to determine the low cov erage zones in the cellular–connected U A V network. This model identiﬁed causes such as two–ray ground reﬂections, diffraction losses and nulls in antenna radiation pattern as the predominant factors for path loss. DRAFT January 24, 2018 17 Path loss and shado w fading statistics for the A G propagation channel presented in this section demonstrated that the U A V ﬂight dynamics, such as the altitude and distance from the ground le vel, are the dominant contributors for the large–scale fading. Therefore, the dev elopment of realistic U A V propagation model requires these parameters to be considered in 3D coordinates. Also, considerable attention is needed for characterizing antenna design and orientations, as this will further improve the U A V communications. In T able III, we summarize the large–scale fad- ing statistics for the A G channel. 2) Small–Scale F ading Statistics: Multipath propagation and small–scale channel character - ization is important to study the impact of the fading channel behavior on U A V communications. In [37], narro wband fading characteristics were analyzed for the lo w ele v ation propagation chan- nel and time–series generator was developed to capture the channel effects. Hence, the v aria- tions of the receiv ed signal voltage ( y ) were statistically analyzed with the PDF of combination of Rician and Log–normal distribution. f ( y ) = y σ 2 p 2 π Σ 2 A Z ∞ a =0 1 a e − (20 log a − M A ) 2 2 Σ 2 A e − ( y 2 + a 2 ) 2 σ 2 I 0 ( y a σ 2 ) da, (1) where M A and Σ are respectively , the mean and standard de viation of the Gaussian distribution for the magnitude of the LOS signal, a and σ , respectiv ely , denotes the magnitude of the LOS and the diffuse multipath components of Ricean distribution, respectively , and I 0 ( . ) is the ze- roth–order modiﬁed Bessel function. Rician and Log–normal compositional models are preferred for the land mobile satellite sys- tems and referred as Loo model [5]. The fading channel variations for these models are studied with the ﬁrst order statistics, such as cumulativ e distribution function (CDF) or probability den- sity function (PDF), and second order statistics such as, average fade duration (AFD) or lev el crossing rate (LCR). Autocorrelation function and power spectral density w as used for the pro- posed time series generator to reproduce channel dynamics. Rician distribution was used in [39] and [49] to analyze the time varying ef fects in the A G propagation channel. In this case, the dominant LOS and diffuse multipath components were January 24, 2018 DRAFT 18 T ABLE III L A R G E – S C A L E FA D I N G S TA T I S T I C S F O R AG C H A N N E L Ref. PL model [38] P L ( dB ) = − 10 log 10 [ 0 . 05 λ 2 h 2 ( d 2 d + r 2 b d 2 r )] − 20 log 10 (1 − e % ) 2 , % = − 0 . 6038 × 0 . 109 v , v ≈ h q 2 λd 2 , h = obstruction height, d 2 = distance between receiver and obstruction, d 2 d = direct–ray distance between recei ver and obstruction, d 2 r = reﬂected–ray distance between recei ver and obstruction, r b = reﬂection coefﬁcient [39] V ertical: P L ( dB ) =      P L ( d 0 ) + 10 α 1 (log 10 d d 0 ) if d < d B P L ( d 0 ) + 10 α 1 (log 10 d d 0 ) + 10 α 2 (log 10 ( d d B )) if d ≥ d B , ( α 1 , σ 1 dB ) = (0.74, 1.23), ( α 2 , σ 2 dB ) = (2.29, 2.15), d B = 9 m Horizontal: P L ( dB ) = P L ( d 0 ) + 10 α (log 10 d d 0 ) , for 20 m: ( α, σ dB , P L ( d 0 ) dB ) =(0.93, 5.5, 77.9), for 30 m: ( α, σ dB , P L ( d 0 ) dB ) =(1.01, 3.9, 74.6) [40] Static UA V: P L ( dB ) = P L ( d 0 ) + 10 α (log 10 d d 0 ) − log 10 4 h h opt + C p + ζ , 4 h = | h g − h opt | , 4 f = ( 4 v c ) f c , ζ ∼ N (0 , σ 2 ) , C p =0 dB, d = 5.6 m to 16.5 m, h g = (1.5 m, 7 cm), ( α, σ dB , P L ( d 0 ) dB ) =2.6471, 3.37, 34.905 (open, 0 km/h), ( α, σ dB , P L ( d 0 ) dB ) =2.7601, 4.8739, 30.4459(suburban, 0 km/h), Mobile U A V: P L ( dB ) = P L ( d 0 ) + 10 α (log 10 d d 0 ) − log 10 4 h h opt + C p + 10 x log 10 ( f c + 4 f f c ) + ζ , , ( α, σ dB , P L ( d 0 ) dB ) =2.6533, 4.02, 34.906(open, 32 km/h), ( α, σ dB , P L ( d 0 ) dB ) =2.8350, 5.3, 30.446(sub- urban, 32 km/h), x = frequency dependent path loss factor and ne gligible at small velocities [41] RS S ( dB m ) = − 95 + 10 log 10 ( K 0 .d − α ) , α =2.34, K 0 = 3 . 6 × 10 − 1 [42] RS S ( dB ) = P t + G + 10 log 10 ( λ 4 πd ) α , P t = 20 dBm, G = 1dB, f c =2.4 GHz, α =2.3 [43] P L ( dB ) = 10 α log 10 ( d ) , α ≈ 2 for beyond 100 m distance [46]- [48] P L ( dB ) = 10 α log 10 ( d ) , α =2.132 (A G), 3.57 (GG) [46], 1.50 (A G) [47], 2.32 (A G), 2.51 (GA), 3.1 (GG) [48] [50] P L ( dB ) = P L ( d 0 ) + 10 α log 10 ( d d 0 ) , d = p d 2 h + d 2 v , P L ( d 0 ) = 46.4 dB, α =2.01 (A G), d h ∈ { 0 , ..., 100 m } , d v = 50 m, d 0 = 1 m [51] RS S ( dB m ) = P rx ( d 0 ) − 10 α log 10 ( d d 0 ) , P rx : receive power at reference distance, α =2.2 (open), 2.5- 2.6(campus) DRAFT January 24, 2018 19 Ref. PL model [52], [53] RS S ( dB m ) = A − 10 αl og 10 ( d ) , α =1.80, A =-37.5 [52], α =1.04 , A =-55.12 [53] [54] Urban: P L ( dB m ) = 89 . 5357 + ( h 3 U AV 10000 + 0 . 0108 h 2 U AV + 0 . 8588 h U AV ) Rural: P L ( dB m ) = 78 . 2186 − 0 . 0013 h 2 U AV − 0 . 0052 h U AV , h U AV ∈ { 0 , ..., 500 m } [55] P L ( dB ) = 20 log ( 4 πd 0 λ ) + X dis + X f req + X hei + X ang , X dis , X f req , X hei , X ang : model parameters in 3D plane [56] P L ( dB ) = α ( h U AV )10 log 10 ( d ) + β ( h U AV ) + ζ , ζ ∼ N (0 , σ ( h U AV )) , α ( h U AV ) = 2.9-2.0 (15-100 m), β ( h U AV ) dB = -1.3-35.3 (15-100 m), σ ( h U AV ) dB = 7.7-3.4 (15-100 m) [57] P L ( dB ) = α 10 log 10 ( d ) + A ( φ − φ 0 ) exp( − φ − φ 0 B ) + η 0 + ζ ζ ∼ N (0 , aφ + σ 0 ) , α = 3.04, A = -23.29, B = 4.14, φ 0 = -3.61, η 0 = 20.70, a = -0.41, σ 0 = 5.86 [65] P L ( dB ) = P tx − 10 log 10 (12 .B W ) − RS RP + G U AV + G B S P tx = maximum transmit power , B W = transmission bandwidth, R S RP = measured reference signal receiv ed power , G U AV = gain of U A V antenna, G B S = gain of base station antenna [68] P L ( dB ) = − 20 log 10 | ν | + 40 log 10 ( d ) − 10 log 10 ( h 2 B S h 2 U AV ) , ν : Kirchoff diffraction parameter α = path loss exponent, RS S = received signal strength, d = separation distance, d 0 = reference distance, d B = breakpoint- distance, C P = foliage loss, σ = standard de viation, h g = height from ground lev el, h opt = optimal height from ground lev el, f c = carrier frequency , h U AV = U A V altitude, h B S base station altitude, 4 f = Doppler shift, K 0 = transmission gain, G = antenna gain, A = y –intercept, λ = wa velength ﬁtted using the PDF of Rician distribution gi ven by f ( y ) = y σ 2 e − ( y 2 + a 2 ) 2 σ 2 I 0 ( y a σ 2 ) , (2) where y ≥ 0 , a and σ denotes the magnitude of the LOS and the diffuse multipath components, respecti vely . Also, the Rician parameter K is deﬁned as: K = a 2 2 σ 2 . (3) In [39], small–scale fading parameters were approximated by the SA GE algorithm and ana- lyzed to study the beha vior of multipath components at different altitude le vels. Also, in [49], multipath propagation effects caused by the ground reﬂection were analyzed by the height–dependent Rician parameter K . January 24, 2018 DRAFT 20 In [40] and [50], the Nakagami m distrib ution was used with PDF gi ven by f ( y ; m, Ω ) = 2 m m Γ( m ) Ω m ( y 2 m − 1 ) e ( − my 2 Ω ) , (4) where Γ( m ) is the Gama function, m and Ω are the Nakagami shape and spread controlling parameters, respecti vely , m = E 2 [ X 2 ] V ar [ X 2 ] (5) Ω = E [ X 2 ] . (6) In [40], the magnitude of individual multipath components was collected for different time delay bins from the multiple channel impulse responses and modeled by the Nakagami m dis- tribution, where the mean ( η ) and standard deviation ( ξ ) of the m parameter were e valuated empirically . As a result, mean is found to be small for both open and suburban areas under the inﬂuence of tree canopy and large variance is observ ed due to thick suburban scattering. In addition, multipath channel characteristics, such as time of arriv al, was modeled as a Poisson process and analyzed by the CDF . Also, time dispersiv e parameters were estimated by Clean’ s algorithm. In this case, frequency dispersiv e parameters were not computed due to the low ve- locity of U A V . In [50], the CDF analysis w as performed to compare the theoretical Rayleigh and Nakagami distribution. It was observed that as the shape parameter m is alw ays greater than 1. Thus, Nakagami m fading channel has the best ﬁt. T ime and frequency dispersi ve parameters were computed in [63] for residential and moun- tainous desert terrain, where cross ambiguity function (referred as scattering function) was opted to estimate channel parameters from the measured channel impulse response. In this work, the mean and v ariance of the po wer delay proﬁle were compared by CDF and it was found that for the mountainous desert scenario, the median RMS delay spread and the Doppler frequency spread are roughly 0.06 µs and 28.96 Hz, respectiv ely . For the residential area, the measured me- dian RMS delay spread and the Doppler frequency spread are approximately 0.03 µs and 28.06 Hz. The RMS delay spread attained in the desert terrain is larger due to the rough mountainous scatters along the ﬂight path than those measured in the residential area. DRAFT January 24, 2018 21 Channel characterization for small–scale fading effects mostly addresses temporal variations and scant ef forts hav e been put to study spatial variation. Also, for most A G propagation cases reported, the Nakagami and Rician distributions seem to effecti vely analyze the fading channel statistics. I V . A NA L Y T I C A L C H A N N E L M O D E L S Analytical channel models are useful for characterizing the propagation beha vior under cer- tain assumptions and parameters. They can predict the performance of communication systems. For the land mobile satellite systems, channel behavior can be analyzed using the multi–state Marko v chain model [4], [5]. For the terrestrial cellular systems, there are three main model- ing approaches: deterministic, stochastic and geometry–based stochastic approach. In the de- terministic technique, en vironmental–speciﬁc parameters are utilized to model the propagation channel. In the stochastic approach, the propagation characteristics are realized by the channel statistics without requiring the speciﬁc location information. In the geometry–based stochastic approach, random scatters are assumed in the en vironment to obtain the spatial–temporal statis- tics in a stochastic manner by applying the deterministic or ray–tracing model. These models can be used for U A V channels. In this section, we will categorize the U A V channel modeling approaches reported in the literature and presents some analytical expressions of them. A. Deterministic In deterministic channel models, en vironmental clutters are placed in the certain layouts. This approach assumes large dimensions of the en vironmental objects in comparison with the wa ve- length, thus not compensating the diffuse scattering. The accuracy of these channel models depends on the en vironment–speciﬁc database which consists of the information related to the terrain topography , the electrical parameters of buildings and other obstruction materials. Deter- ministic models can also be realized by the ray–tracing simulation software, which can depict the realistic behavior of the EM wa ve propagation and simulate po wer loss and shadowing ef- fects. January 24, 2018 DRAFT 22 In [6], 3D ray–tracing was performed to characterize the A G propagation in the suburban en- vironment for the channel between cellular base station and airborne U A V . Also, the well–known macro cell terrestrial channel models were tested with the boundary conditions in order to de- termine the applicability of lo w altitude aerial platforms for providing cellular co verage. In [7] and [8], analytical propagation models hav e been studied for the A G channel characterization in an urban en vironment for the frequency ranging from 200 MHz to 5 GHz and the aerial altitude from 100 to 2000 m. In [7], the average path loss and shado wing statistics were examined as the function of elev ation angle and the aerial altitude through 3D ray–tracing simulations. The authors hav e provided analytical path loss expressions. Also, the shadow fading was ﬁtted with the log–normal distribution with the standard deviation dependent on the ele v ation angle. In addition, propagation conditions were determined from the simulation results and categorized as, LOS, NLOS (non-LOS) and OLOS (obstructed LOS) channels. The work in [8] utilized knife–edge diffracti on theory to model the LOS probability , which considered the statistical parameters to account for height, size and cov erage area of buildings in the simulation en viron- ment. In [9]– [11], en vironmental topography was realized with the statistical parameters recom- mended by the International T elecommunication Union (ITU–R). In [9], a generic path loss model in low altitude platform was proposed, where the channel model parameters were esti- mated by the 3D ray–tracing at 700 MHz, 2000 MHz and 5800 MHz. In this work, the A G channel conditions fa v oring LOS and NLOS propagations were grouped distinctly and analyzed with the group occurrence probability as the conditional PDF . Simulation results demonstrated that the impact of ele vation angle w as signiﬁcant on the excess path loss. In [10], the closed–form expressions was formulated for predicting the coverage footprint from the aerial platform in terms of the maximum cell radius and the optimal altitude. In this study , the free space path loss model w as extended with the excessi ve attenuation factor and cor - responds to LOS and NLOS propagation conditions. This was extended in [11] to provide the analytical frame work for optimization of the av erage radio cov erage probability and the max- imum transmission rate to achie ve the required quality of service. Some of the deterministic U A V channel models are reported in T able IV DRAFT January 24, 2018 23 B. Stoc hastic Channel Model For the U A V communication systems, stochastic based channel models can be designed us- ing the tapped delay line (TDL) system with dif ferent numbers of taps, each of which can ac- commodate fading statistics of the multipath components deriv ed from the channel impulse response. The accuracy of these model depends on the estimation of stationary interv al in the non–stationary U A V channel. In [12]– [14], wideband stochastic channel models were proposed from the data collected in different en vironments, using the estimated stationary interv al of 15 m at the C band. For the over water settings in [12], the A G channel employed the TDL model to characterize the two–ray propagation plus an intermittent multipath component as the third ray . In this work, the authors hav e argued that the statistics for LOS and reﬂected components can be analyzed by either curved earth two ray (CE2R) or ﬂat earth two ray (FE2R) model. The probability of the existence of the intermittent multipath component was estimated by the exponential distri- bution as a function of link distance. The TDL model with nine taps hav e been proposed for the mountainous terrain [13] and an urban en vironment [14]. In both of these studies, sev en in- termittent multipath components were considered and the probability of e xistence, excess delay and duration of intermittent multipath components were modeled as the linear function of link range. In [69] and [70], stochastic model was de veloped with the narro wband assumption to charac- terize the aeronautical A G channel. In [69], the stochastic model was designed for characterizing the A G propagation in terms of transmission coef ﬁcients assuming that the quadrature compo- nents reﬂected from the ground surface can be modeled as a zero–mean Gaussian process. Also, Doppler spectrum analysis was performed for the diffuse multipath components. In [70], the proposed model was de veloped with the TDL system having both LOS and NLOS taps, where the amplitude attenuation and the multipath delay of NLOS components were assumed to be Rayleigh distributed and Gaussian random process, respectiv ely , while the phase shift was uni- formly distributed. Further , the Doppler frequency shift was characterized as the time varying random process and the channel stationarity interval was not computed, but the fading statistics January 24, 2018 DRAFT 24 T ABLE IV D E T E R M I N I S T I C M O D E L S Ref. Analytical Model Parameters [7] Path loss: P L ( dB ) =            − 0 . 58 + 0 . 549 e (90 − φ ) 24 LOS channel η 0 − η 1 e − (90 − φ ) ν NLOS channel ι 0 − ι 1 e (90 − φ ) ω OLOS channel Shadow f ading: σ ( dB ) = ρ (90 − θ ) γ φ : elev ation angle, for 200 MHz: ( η 0 , η 1 , ν )= (9.08, 6.40, 12.01), ( ι 0 , ι 1 , ω )= (2.11, 0.41, 22.07), for 5000 MHz: ( η 0 , η 1 , ν )= (20.43, 14.60, 10.50), ( ι 0 , ι 1 , ω )= (6.23, 0.4787, 22.65), LOS channel at 200 MHz and 100 m altitude: ( ρ, γ )= (0.0143, 0.9941), NLOS chan- nel at 200 MHz & 5000 MHz: ( ρ, γ )= (0.7489, 0.4638) & (2.7940, 0.2259), OLOS channel at 200 MHz & 5000 MHz: ( ρ, γ )= (0.3334, 0.3967) & (0.8937, 0.3713) [8] LOS probability in a street: P LOS =      1 − S c sin ϑ W s , 0 < S c < W s / sin ϑ 0 , S c > W s / sin ϑ φ : ele vation angle, ϑ : street angle, ϑ c : critical street angle, W s : street width, S c : critical distance be- tween ground station to adjacent buildings, H = building height, h g : ground station height, W e : estimated street width, λ : wav elength, W s = 15m, ϑ = 90 ◦ , ∆ H = H − h g , H = 11.71m, h g = 15m, W e =44.2m LOS probability in an area: S c > ∆ H cot φ + 0 . 16 λ cos φ + √ (0 . 16 λ cos φ ) 2 +0 . 32 λ ∆ H sin φ sin 2 φ , P LOS = 2 π [ ϑ c − S c (1 − cos ϑ c ) W e ] sin ϑ c =      W e S c , W e ≤ S c 0 , otherwise [9] Path loss: P L ( dB ) = 20 log ( 4 h sin φ ) + 20 log( f M H z ) − 27 . 55 h U AV : U A V altitude, 4 h = h U AV − h g , h U AV = 200m, h g = 1.5 m, f M H z = 700, 2000, 5800 LOS probability: P LOS = a ( φ − φ 0 ) b , φ 0 = 15 ◦ , For 700 MHz: suburban( a = 0 . 77 , b = 0 . 05 ), urban( a = 0 . 63 , b = 0 . 09 ), dense urban( a = 0 . 37 , b = 0 . 21 ), high–rise urban( a = 0 . 06 , b = 0 . 58 ) [10] Path loss: P L ( dB ) =      20 log( 4 πf c d c ) + ε LOS LOS channel 20 log( 4 πf c d c ) + ε N LOS NLOS channel d = p h U AV 2 + r 2 , r = cell radius, f c = 2000 M H z : suburban( ε LOS = 0 . 1 , ε N LOS = 21 ), urban( ε LOS = 1 . 0 , ε N LOS = 20 ), dense urban( ε LOS = 1 . 6 , ε N LOS = 23 ), high–rise urban( ε LOS = 2 . 3 , ε N LOS = 34 ) DRAFT January 24, 2018 25 T ABLE V T D L M O D E L S Ref. TDL model [12] h ( τ , t ) = h 2 − ray ( τ , t ) + w 3 ( t ) A 3 ( t ) exp( − j ϕ 3 ( t )) δ ( τ − τ 3 ( t )) , h 2 − ray denotes FE2R or CE2R model, w 3 ( t ) ∈ { 0 , 1 } represents presence/absence of third–ray and modeled as p ( d ) = ae bd , A 3 is the amplitude of third–ray and modeled by the Gaussian distribution, ϕ 3 ∈ { 0 , 2 π } is the uniformly randomly distributed phase of third–ray , τ 3 is the excess delay of third–ray and modeled as p ( τ 3 ) = 1 µ e − ( τ 3 − 100 /µ ) , ( a, b ) =(0.17,-0.25) ov er sea water and (0.03,-0.15) over freshwater , µ = 17 ns, 6 ns ≤ τ 3 ≤ 7 ns , d is the link distance [13], [14] h ( τ , t ) = A 1 ( t ) δ ( τ − τ 1 ( t )) + A 2 ( t ) exp( − j ϕ 2 ( t )) δ ( τ − τ 2 ( t )) + P 9 L =3 w L ( t ) A L ( t ) exp( − j ϕ L ( t )) δ ( τ − τ L ( t )) A , ϕ and τ denotes amplitude, phase and excess delay , respectiv ely , subscripts 1, 2 and L represents LOS, reﬂected and L th intermittent multipath components, respectively , variations of w L and τ L are modeled as a linear function of link range , ϕ L ∈ { 0 , 2 π } is the the uniformly randomly distributed phase, 10 log( A 2 L A 2 1 ) represents relativ e power of intermittent components and follo ws a Gaussian distribution [70] y ( t ) = A 1 ( t ) cos[2 π { f c + 4 f } ( t − τ 1 ( t ))] + P N L =2 A L ( t ) cos[2 π { f c + 4 f } ( t − τ L ( t )) + ϕ L ( t )] + n ( t ) A 1 is the amplitude of LOS path, A L represents amplitude of NLOS paths and assumed as Rayleigh random process, ϕ L ∈ {− π , π } is the phase shift of NLOS paths and modeled as uniform random process, 4 f denotes Doppler frequency shift and modeled as time–variant random process, τ 1 = 25 µ s, τ L is excess delay of NLOS components and modeled as the Gaussian random process with mean and standard deviation of 30 µ s and 5 µ s, respectiv ely , n ( t ) is white Gaussian noise were assumed to be constant for the random time duration. In T able V , we have presented the channel response from the TDL models reported in this paper . C. Geometry–based Stochastic Channel Model Geometry–based stochastic modeling approach obtains the spatial–temporal channel char- acteristics with stochastic output in a 3D geometric simulated en vironment. The accuracy of this model is dependent of the physical en vironment surrounded with scatters following certain probability distrib utions. The geometric based channel model for the analysis and simulation of the A G radio communication was proposed in [15]. It characterized the multipath propagation in a cluttered en vironment around the ground station conﬁned within a virtual 3D ellipsoidal geometry to analytically ev aluate delay , gain, phase and angle of arriv al (A O A) of individual multipath components. The path loss can be determined using the log–distance model between January 24, 2018 DRAFT 26 the airborne platform and the clutters. Therefore, the proposed model is equally applicable to de- termine both narro wband and wideband channel statistics and well suited for designing antenna di versity system and antenna arrays. This was extended in [16] to theoretically estimate the MIMO performance for the low altitude A G channel and also characterize the propagation loss for LOS and multipath components using the log–distance path loss model with the log–normal shado w fading. In this model, the small–scale spatial fading was modeled by the Ricean distri- bution to analyze the scattering of the multipath components. Furthermore, the probability of error was simulated for SISO system and compared with a 2 × 2 space time block coding and a 2 × 2 spatial multiplexing gain using maximum likelihood detection. In [17], 3D A G propagation model was proposed for the dense scattering environment considering lo w altitude platform. The model was deri ved for a direction of arri v al and the delay dependent Doppler spectrum with the approximation of linear distribution of the scattering point. In this work, the analytical results were compared with the terrain based digital elev ation model simulation results and found that the terrain morphology af fects the Doppler–delay spread spectrum. In [18], a realistic 3D geometric–based stochastic model has been de veloped for the A G com- munication between an airborne platform and the base station as an elev ated plane. The pro- posed model considered scattering points as uniformly distrib uted around the base station. In this study , the spatial characteristics were analyzed with the closed–form analytical expressions. In [19], geometric–based stochastic approach has been utilized for U A V channel modeling to analytically characterize a 2 × 2 MIMO enabled A G propagation in 3D plane. In this case, the model w as de veloped with the assumption that the ground scatters are distrib uted on the cylin- drical surface and scatter free airborne en vironment. Based on the proposed model, analytical expressions were used to study the impact of ele vation angle and direction of U A V movement on the space time correlation function in a non–isotropic en vironment. Some analytical e xpressions to determine A O A using geometry–based model are gi ven in T able VI. V . I M P O RTA N T I S S U E S A. Airfr ame Shadowing In aeronautical communication, the radio path between aircraft and ground control station may be blocked by aircraft structure, such as wings, fuselage or engine. Also, during ﬂight DRAFT January 24, 2018 27 T ABLE VI G E O M E T RY – BA S E D S T O C H A S T I C M O D E L Ref. Geometry–based stochastic model [15] The PDF of A OA as the function of ele vation angle ( φ ) around the ground receiver: f ( φ ) = ( x 2 a x 2 a − x 2 b ) − 1 2 πγ ( x a √ x 2 a − x 2 b − cos φ ) 2 , x a and x b are subsequently the major and minor axis of the planar elliptical scattering surface, γ = x a √ ( x 2 a − x 2 b ) [ x 2 a x 2 a − x 2 b − 1] 1 2 [18] The PDF of A OA seen at the airborne platform: f (Ψ ap , φ ap ) = ( l 3 ap,max − l 3 ap,min ) cos φ ap 3 V , Ψ ap and φ ap are, respectiv ely , the azimuth and the ele vation angle observed from the airborne platform, l ap,max and l ap,min are the distance between the U A V and the farthest and nearest scatter point, respectiv ely . The PDF of A OA seen around the ele vated ground plane: f (Ψ bs , φ bs ) = ( l 3 bs,max − l 3 bs,min ) cos φ bs 3 V , Ψ bs and φ bs are, respectively , the azimuth and the elev ation angle observed from the base station, l bs,max and l bs,min are the distance between the base station and the farthest and nearest scatterer point, respectiv ely , and V is the volume of the scattering region [19] The von Mises PDF of A O A as the function of azimuth angle: f (Ψ) = e k cos(Ψ − Ψ µ ) 2 πI 0 ( k ) , − π Ψ ≤ π , k is a spreading control parameter , Ψ µ ∈ [ − π , π ] is the mean angle of the distribution of scatterers in a 2D plane, I 0 ( . ) is the modiﬁed Bessel function of the zeroth–order, k =3, Ψ µ = π The cosine PDF of A OA as the function of ele vation angle: f ( φ ) = π 4 φ m cos( π 2 φ − φ µ φ m ) , mean angle φ µ = π 6 and variance φ m = π 4 maneuvering or banking turns, the direct LOS path may se vered and thus, induced shadowing. In the context of U A V channel characterization, airframe shado wing is still unexplored, as most of the measurement campaigns pertinent to study this phenomena are initiated with manned air- crafts in high altitude. Therefore, the characterization of airframe shado wing with multi–rotor U A Vs in lo w altitude is an interesting topic. Here, we can only have airframe shadowing char- acterization with manned aircrafts. In [20], channel measurements were extracted for the communication link between aircraft and satellite, where shadowing effect was induced by the forced aircraft maneuvering. Charac- terization of the A G channel in C band w as performed in [21], where the propagation signal w as obstructed by the right banking turn when the aircraft mo ving in the circular ﬂight track and led to the weakest recei ved signal. Similarly , strongest signal was detected during the left banking turn. Related measurements were conducted in [22], where CDF analysis of the received signal January 24, 2018 DRAFT 28 po wer during the circular ﬂight track demonstrated that the airframe shado wing can be modeled by the Gaussian distribution. The authors hav e observed that the shado wing effects were sub- stantial in the circular ﬂight track than in linear ﬂight proﬁle. In [23], airframe shadowing was reported due to wings and engine of the commercial A320 aircraft, where shado wing statistics were simulated by the ﬁnite dif ference time domain approach. Empirical airframe shadowing model was proposed in [24], for analyzing shadowing loss and duration. In this study , aircraft follo wed ov al ﬂight route and found that shadowing statistics were disjoint from the ground en- vironment and link distance. B. Stationary Interval One of the most important characteristics that distinguish U A V communication from the con- ventional terrestrial wireless systems is the non–stationarity in U A V channels, when the WSSUS assumption is violated. Therefore, wideband frequency–dispersi ve channel statistics can possess any signiﬁcance within the stationary interval of the non–stationary U A V channel. No compre- hensi ve study is a vailable in the literature that addresses channel non–stationarity for the U A V propagation channel in lo w altitude platform. Therefore, estimation of the stationary interval is a contemporary research topic. Efforts to characterize the A G channel with stationarity interval was performed in [60], using manned aircraft at high altitude platform. In this study , stationary interv al was computed for the wideband measurements using temporal PDP(power delay pro- ﬁle) correlation coef ﬁcient method, whereas, spatial correlation collinearity was considered for narro wband measurements. The estimated stationary interv al from both of these methods are approximately 15 m or 250 λ at C band with 50 MHz bandwidth. C. Diversity Gain Di versity techniques are beneﬁcial to enhance the reliability of the communication systems, particularly when deep fades dominate. Div ersity possibilities hav e been mostly exploited in MIMO airborne communications with manned aircrafts. For example, in [25], a 4 × 4 MIMO enabled OFDM system was used to increase the average throughput by 2 times and the range extension by 1.6 times in comparison to a SISO system. In [26], multiple helicopter mounted antennas were utilized to achiev e the signal–to–noise (SNR) gain of approximately 13 dB. In DRAFT January 24, 2018 29 [28], the spatial multiplexing gain was achie ved with a 2 × 2 MIMO conﬁguration and as a consequence the throughput gain was enhanced up to 8 times for most of the ﬂight route. In the context of U A V communications, there are fe w measurement campaigns on the effect of multiple antenna elements. In [61] and [62], the A G channel characterization was initiated with a 1 × 4 antenna conﬁguration. In this w ork, carrier–to–noise ratio (CNR) gain was compared for the common combining strategies such as selection, equal–gain and maximal ratio combining (MRC) with dif ferent antenna elements and observed that the div ersity gain achiev ed with the MRC method using four antenna elements is approximately 4 dB greater in an urban en viron- ment than in a wooded area under similar circumstances. In [64], the performance of multiple receiv er and transmitter nodes was ev aluated by the cor- relation coefﬁcient. In this case, the packet deliv ery rate was boosted by 25% on average due to the poor correlation at the multiple receiv er nodes in a 1 × 4 conﬁguration and by 37% with the selection di versity using three transmitters in a 3 × 4 setup. Measurement analysis of a 4 × 4 MIMO channel in [71] rev ealed that despite of the sparse multipath en vironment, poor spatial correlation provides the signiﬁcant capacity gain due to the planar wav efronts generated by near–ﬁeld reﬂections at the ground recei ver side. It is argued that more div ersity gain could be achiev ed with the robust airborne platform, constructed from the con ventional aircraft mate- rials. In this study , ground spatial characteristics were estimated by the Cosine Hermitian angle. The div ersity gain achiev ed by multiple antennas are dependent on the number of transmitter and receiv er antennas and the operating en vironment of both the U A V and the ground station. Not enough experimental setups were designed to comprehensiv ely characterize the viability of these systems. Substantial research works are still required to fully recognize the beneﬁts of MIMO technology for both A G and AA propagation. V I . F U T U R E R E S E A R C H C H A L L E N G E S In this section, we will discuss some future research challenges for characterizing the U A V channel with measurement campaigns and in the de velopment of realistic U A V channel model: January 24, 2018 DRAFT 30 • T o provide seamless coverage with U A V communication in all circumstances, measurement campaigns are required to inv estigate the U A V channel in dense urban en vironment and in the metropolitan cities with consent from the local ci vil aviation re gulatory bodies. • USRP platforms can pro vide more ﬂe xibility to initiate U A V measurement campaigns with wideband frequencies and lo w po wer consumption. This also allo ws USRP to test dif ferent wireless communication protocols such as multi–carrier and MIMO system to be used in U A V communication. • The AA channel characterization is required to study the consequences of Doppler shift experienced by multiple U A Vs cruising with dif ferent velocities. • Airframe shadowing has not receiv ed commensurate le vel of attention for small size rota- tory U A Vs, Therefore, measurement campaigns are required to study this phenomenon for both A G channel in the single–hop netw ork and AA propagation in the multi–hop network. In addition, ray–tracing can be used to probe the airframe shado wing, as CAD tools are ca- pable of incorporating U A V shape, metallic properties and different maneuvering positions. • Estimation of the stationary interval is of paramount importance for characterizing the A G channel regarding wideband frequency–selecti ve parameters. Therefore, it would be in- teresting to estimate channel parameters with spectral di ver gence [72] and e volutionary spectrum [73] methods. V I I . C O N C L U S I O N S This paper has provided a comprehensi ve surve y of the U A V channel characterization with measurement campaigns and statistical channel models. W e have categorized the U A V channel measurement campaigns in low altitude platform based on the narro wband or wideband chan- nel sounder , low–cost and low–po wer channel sounding solution, and widely deployed ground infrastructure. W e ha ve also re vie wed empirical channel models for characterizing A G and AA propagation channel. Then we hav e classiﬁed the U A V channel modeling approaches as deterministic, stochastic and geometric–stochastic models. Further , we hav e e xamined some challenging issues in the practicability of U A V communications related to airframe shadowing, the channel non–stationarity , and div ersity techniques. Finally we have presented some future DRAFT January 24, 2018 31 research challenges which will be helpful to pro vide further insight of the U A V channel charac- terization for launching future measurement campaigns and proposing pragmatic framew ork for the ef fectiv e U A V channel models. R E F E R E N C E S [1] Radio T echnical Commission for Aeronautics (R TCA), SC-228, “Minimum Operational Performance Standards for Un- manned Aircraft Systems”, May 2013, av ailable online at http://www .rtca.org/content/sc-228. [2] Radio T echnical Commission for Aeronautics (R TCA), “Drone Advisory Committee (DA C)”, 2016, av ailable online at http://www .rtca.org/content/drone-advisory-committee. [3] National Aeronautics and Space Administration, Oct. 2014, available online at http://www .aeronautics.nasa.gov/isrp/uas/index. html. [4] A. Panagopoulos, P .D.M. Arapoglou and P . Cottis, ”Satellite communications at KU, KA, and V bands: Propagation impairments and mitigation techniques, ” IEEE Commun. Surv . T uts. , vol. 6, no. 3, pp. 2-14, 2004. [5] P . Chini, G. Giambene and S. Kota, ”A survey on mobile satellite systems, ” Int. J. Satell. Commun. Networking , vol. 28, no. 1, pp. 29-57, 2009. [6] K. Daniel, M. Putzke, B. Dusza and C. W ietfeld, ”Three dimensional channel characterization for low altitude aerial vehicles, ” in Pr oc. 7th Int. Symp. W ir eless Commun. Syst (ISWCS’10) , Y ork, UK, Sept. 2010, pp. 756-760. [7] Q. Feng, J. McGeehan, E. K. T ameh and A. R. Nix, ”Path Loss Models for Air -to-Ground Radio Channels in Urban En vironments, ” in Pr oc. IEEE V eh. T echnol Conf. (VTC-Spring’06) , Melbourne, Australia, May 2006, pp. 2901-2905. [8] Q. Feng, E. K. T ameh, A. R. Nix and J. McGeehan, ”Modelling the Likelihood of Line-of-Sight for Air-to-Ground Radio Propagation in Urban En vironments, ” in Pr oc. IEEE Global Commun. Conf. (GLOBECOM’06) , San Francisco, USA, Dec. 2006, pp. 1-5. [9] A. Al-Hourani, S. Kandeepan and A. Jamalipour, ”Modeling Air -to-Ground Path Loss for Low Altitude Platforms in Urban En vironments, ” in Pr oc. IEEE Global Commun. Conf. (GLOBECOM’14) , Austin, USA, Dec. 2014, pp. 2898-2904. [10] A. Al-Hourani, S. Kandeepan and S. Lardner , ”Optimal LAP Altitude for Maximum Cov erage, ” IEEE W ireless Commun. Lett. , vol. 3, no. 6, pp. 569-572, 2014. [11] A. Al-Hourani, S. Chandrasekharan, G. Kaandorp, W . Glenn, A. Jamalipour and S. Kandeepan, ”Cov erage and Rate Analysis of Aerial Base Stations, ” IEEE T rans. Aer osp. Electr on. Syst. , vol. 52, no. 6, pp. 3077-3081, 2016. [12] D.W . Matolak and R. Sun, ”Air–Ground Channel Characterization for Unmanned Aircraft Systems–Part I: Methods Mea- surements and Models for Over -W ater Settings, ” IEEE T rans. V eh. T echnol. , v ol. 66, no. 1, pp. 26-44, 2017. [13] R. Sun and D.W . Matolak, ”Air–Ground Channel Characterization for Unmanned Aircraft Systems–Part II: Hilly and Mountainous Settings, ” IEEE T rans. V eh. T echnol. , v ol. 66, no. 3, pp. 1913-1925, 2017. [14] D.W . Matolak and R. Sun, ”Air–Ground Channel Characterization for Unmanned Aircraft Systems–Part III: The Suburban and Near-Urban En vironments, ” IEEE T rans. V eh. T echnol. , v ol. 66, no. 8, pp. 6607-6618, 2017. [15] W . Newhall and J. Reed, ”A geometric air–to–ground radio channel model, ” in Proc. IEEE Mil. Commun. Conf. (MIL- COM”02) , Anaheim, USA, Oct. 2002, pp. 632-636. [16] M. W entz and M. Stojanovic, ”A MIMO Radio Channel Model for Low-Altitude Air-to-Ground Communication Systems, ” in Pr oc. IEEE V eh. T echnol Conf. (VTC–F all’15) , Boston, USA, Sept. 2015, pp. 1-6. [17] M. Ibrahim and H. Arslan, ”Air–Ground Doppler–Delay Spread Spectrum for dense scattering en vironments, ” in Pr oc. IEEE Mil. Commun. Conf. (MILCOM’15) , T ampa, USA, Oct. 2015, pp. 1661-1666. January 24, 2018 DRAFT 32 [18] S. Gulfam, S. Nawaz, M. Patwary and M. Maguid, ”On the Spatial Characterization of 3-D Air -to-Ground Radio Commu- nication Channels, ” in Pr oc. IEEE Int. Conf. Commun. (ICC’15) , London, UK, June 2015, pp. 2924-2930. [19] L. Zeng, X. Cheng, C.X. W ang and X. Y in, ”A 3D Geometry-based Stochastic Channel Model for UA V–MIMO Channels, ” in Pr oc. IEEE W ir eless Commun. Netw . Conf. (WCNC’17) , San Francisco, USA, March 2017, pp. 1-5. [20] M. Holzbock and C. Senninger , ”An aeronautical multimedia service demonstration at high frequencies, ” IEEE T rans. Multimedia , vol. 6, no. 4, pp. 20–29, 1999. [21] Y . Lee, Y . Meng and Y .H. Heng, ”Experimental characterizations of an air to land channel over sea surface in C band, ” in Pr oc. 29th USRI General Assembly , 2008, pp. 2-5. [22] Y .S. Meng and Y .H. Lee, ”Study of shadowing effect by aircraft maneuvering, ” Int. J. Electron. Commun. , vol. 66, no. 1, pp. 7-11, 2012. [23] J. Kunisch, I.de la T orre, A. Winkelma nn, M. Eube and T . Fuss, ”W ideband time–variant air–to–ground radio channel measurements at 5 GHz, ” in Pr oc. 5th Eur . Conf. Antennas Pr opag. (EuCAP’11) , Rome, Italy , April 2011, pp. 1386-1390. [24] R.Sun and D.W . Matolak, ”Air–Ground Channel Characterization for Unmanned Aircraft Systems Part–IV : Airframe Shadowing, ” IEEE Tr ans. V eh. T echnol. , v ol. 66, no. 9, pp. 7643-7652, 2017. [25] J. Chen, B. Daneshrad and W . Zhu, ”MIMO Performance Evaluation for Airborne Wireless Communication Systems, ” in Pr oc. IEEE Mil. Commun. Conf. (MILCOM’11) , Baltimore, USA, Nov . 2011, pp. 1827-1832. [26] M. Rice and M. Jensen, ”Multipath Propagation for Helicopter-to-Ground MIMO Links, ” in Pr oc. IEEE Mil. Commun. Conf. (MILCOM’11) , Baltimore, USA, Nov . 2011, pp. 447-452. [27] M. Rice and M. Saquib, ”MIMO Equalization for Helicopter-to-Ground Communications, ” in Proc. IEEE Mil. Commun. Conf. (MILCOM’11) , Baltimore, USA, Nov . 2011, pp. 501-506. [28] Y . Jiang, A. Tiwari, M. Rachid and B. Daneshrad, ”MIMO for airborne communications [Industry Perspectiv es], ” IEEE W ireless Commun. , vol. 21, no. 5, pp. 4-6, 2014. [29] W . V ergara, J. Lev atich and T . Carroll, ”VHF air–ground propagation far beyond the horizon and tropospheric stability , ” IRE Antennas Propa g. , vol. 10, no. 5, pp. 608–621, 1962. [30] K. Chamberlin, ”The effect of tree cover on air–ground, VHF propagation path loss, ” IEEE T rans. Commun. , vol. 34, no. 9, pp. 958-962, 1986. [31] J.R. Child, ”Air–to–ground propagation at 900 MHz, ” in Proc. IEEE V eh. T echnol Conf.(VTC–F all’85) , Colorado, USA, May 1985, pp. 73-80. [32] M. Rice and R. Dye, ”Narrowband channel model for aeronautical telemetry , ” IEEE T rans. Aer osp. Electr on. Syst. , vol. 36, no. 4, pp. 1371-1376, 2000. [33] W .G. Newhall, R. Mostafa, C. Dietrich, C.R. Anderson, G. Joshi and J.H. Reed, ”Wideband air-to-ground radio chan- nel measurements using an antenna array at 2 GHz for low-altitude operations, ” in in Pr oc. IEEE Mil. Commun. Conf. (MILCOM’03) , Boston, USA, Oct. 2003, pp. 1422-1427. [34] M. Rice, A. Davis and C. Bettweiser , ”W ideband channel model for aeronautical telemetry , ” IEEE T rans. Aerosp. Electron. Syst. , vol. 40, no. 1, pp. 57-69, 2004. [35] US Department of T ransportation Federal A viation Authority , ”Inte gration of Civil Unmanned Air- craft Systems (U AS) in the National Airspace System (N AS) Roadmap, ” 2013, av ailable online at http://www .faa.gov/uas/media/uas roadmap 2013.pdf. [36] Civil A viation Safety Authority (CASA), “Flying drones or model aircraft recreationally”, available online at https://www .casa.gov .au/modelaircraft. [37] M. Simunek, F .P . Fontan and P . Pechac, ”The U A V Low Elev ation Propagation Channel in Urban Areas: Statistical Anal- ysis and T ime-Series Generator , ” IEEE T rans. Antennas Propa g. , vol. 61, no. 7, pp. 3850-3858, 2013. DRAFT January 24, 2018 33 [38] M. Simunek, P . Pechac and F .P . Fontan, ”Excess Loss Model for Low Elev ation Links in Urban Areas for UA Vs, ” Radio engineering , vol. 20, no. 3, pp. 561-568, 2011. [39] X. Cai, A.G-. Plaza, D. Alonso, L. Zhang, C.B. Rodriguez, A.P . Y uste and X. Y in, ”Low altitude UA V propagation channel modelling, ” in Pr oc. 11th Eur . Conf. Antennas Pr opag (EuCAP’17) , Paris, France, March 2017, pp. 1443-1447. [40] W . Khawaja, I. Guvenc and D. Matolak, ”UWB Channel Sounding and Modeling for UA V Air-to-Ground Propagation Channels, ” in Pr oc. IEEE Global Commun. Conf. (GLOBECOM’16) , W ashington, USA, Dec. 2016, pp. 1-7. [41] E.W . Frew and T .X. Brown, ”Airborne Communication Networks for Small Unmanned Aircraft Systems, ” Pr oceedings of the IEEE , vol. 96, no. 12, pp. 2008-2027, 2008. [42] N. Goddemeier , S. Rohde and C. Wietfeld, ”Experimental V alidation of RSS Dri ven UA V Mobility Behaviors in IEEE 802.11s Networks, ” in Pr oc. IEEE Global Commun. Conf. (GLOBECOM’12) , Anaheim, USA, Dec. 2012, pp. 1550-1555. [43] E. Y anmaz, S. Hayat, J. Scherer and C. Bettstetter, ”Experimental Performance Analysis of T wo-Hop Aerial 802.11 Net- works, ”in Pr oc. IEEE W ireless Commun. Netw . Conf . (WCNC’14) , Istanbul, T urkey , April 2014, pp. 3118-3123. [44] T . Brown, B. Argro w , C. Dixon, S. Doshi, R. Thekkekunnel and D. Henkel, ”Ad-hoc UA V ground network (Augnet), ” in Pr oc. AIAA Unmanned Unlimited T ech. Conf. , 2004, pp. 29-39. [45] S. Hayat, E. Y anmaz and C. Bettstetter , ”Experimental Analysis of Multipoint-to-Point UA V Communications with IEEE 802.11n and 802.11ac, ” in Pr oc. 26th IEEE Annual Intl. Symp. P ersonal Indoor Mobile Radio Commun. (PIMRC’15) , Hong K ong, 2015. pp. 1991-1996. [46] J. Allred, A. B. Hasan, S. Panichsakul, W . Pisano, P . Gray and J. Huang, ”SensorFlock: An Airborne Wireless Sensor Network of Micro-Air V ehicle, ” in Pr oc. ACM SenSys , 2007, pp. 117–129. [47] A. Shaw and K. Mohseni, ”A Fluid Dynamic Based Coordination of a W ireless Sensor Network of Unmanned Aerial V ehicles: 3-D Simulation and W ireless Communication Characterization, ” IEEE Sensors Journal , vol. 11, no. 3, pp. 722- 736, 2011. [48] N. Ahmed, S. Kanhere and S. Jha, ”On the Importance of Link Characterization for Aerial Wireless Sensor Network, ” IEEE Commun. Mag. , vol. 54, no. 5, pp. 52-57, 2016. [49] N. Goddemeier and C. W ietfeld, ”Inv estigation of Air–to–Air Channel Characteristics and a U A V Speciﬁc Extension to the Rice Model, ” in Pr oc. IEEE Global Commun. Conf. (GLOBECOM’15) ,San Diego, USA, Dec. 2015, pp. 1-5. [50] E. Y anmaz, R. Kuschnig and C. Bettstetter , ”Achieving Air–Ground Communications in 802.11 Networks with Three- Dimensional Aerial Mobility , ” in Pr oc. IEEE INFOCOM , T urin, Italy , April 2013, pp. 120-124. [51] E. Y anmaz, R. Kuschnig and C. Bettstetter, ”Channel Measurements ov er 802.11a–based U A V–to–Ground Links, ” in Proc. IEEE Global Commun. Conf. (GLOBECOM’11) , Houston, USA, Dec. 2011, pp. 1280-1284. [52] C.M. Cheng, P .H. Hsiao, H. K ung and D. Vlah, ”Performance measurement of 802.11a wireless links from U A V to ground nodes with various antenna orientations, ” in Pr oc. Intl. Conf. Computer Comm. and Networks (ICCCN’06) , Arlington, USA, Oct. 2006, pp. 303-308. [53] D. Hague, H. Kung and B. Suter, ”Field Experimentation of Cots-Based UA V Networking, ” in Pr oc. IEEE Mil. Commun. Conf. (MILCOM’06) , W ashington, USA, Oct. 2006, pp. 1-7. [54] N. Goddemeier , K. Daniel and C. W ietfeld, ”Cov erage Evaluation of Wireless Networks for Unmanned Aerial Systems, ” in Pr oc. IEEE Global Commun. Conf. (GLOBECOM’10) , Miami, USA, Dec. 2010, pp. 1760-1765. [55] T . T av ares, P . Sebastiao, N. Souto, F . Cercas, M. Ribeiro, A. Correia and F .J. V elez, ”Generalized LUI propagation model for U A Vs communications using terrestrial cellular networks, ” in Proc. IEEE V eh. T echnol Conf. (VTC–F all’15) , Boston, USA, Sept. 2015, pp. 1-6. [56] R. Amorim, H. Nguyen, P . Mogensen, I.Z. K ovacs, J. Wig ard and T .B. Sorensen, ”Radio Channel Modeling for U A V Communication Over Cellular Networks, ” IEEE Wir eless Commun. Lett. , vol. 6, no. 4, pp. 514-517, 2017. January 24, 2018 DRAFT 34 [57] A.Al-Hourani and K. Gomez, ”Modeling Cellular–to–UA V Path-Loss for Suburban En vironments, ” IEEE W ireless Com- mun. Lett. , vol. pp, no.99, 2017. [58] L. Gupta, R. Jain and G. V aszkun, ”Surve y of Important Issues in UA V Communication Networks, ” IEEE Commun. Sur . T ut. , vol. 18, no. 2, pp. 1123-1152, 2016. [59] S. Hayat, E. Y anmaz, R. Muzaffar , ”Survey on Unmanned Aerial V ehicle Networks for Civil Applications: A Communi- cations V iewpoint, ” IEEE Commun. Sur . T ut. vol. 18, no. 4, pp. 2624-2661, 2016. [60] D.W . Matolak, ”Air-Ground Channels and Models: Comprehensive Revie w and Considerations for Unmanned Aircraft Systems, ” in Pr oc. IEEE Aero. Conf. , MT , USA, March 2012, pp. 1-17. [61] M. Simunek, F .P . Fontan, P . Pechac and F .J.D. Otero, ”Space Div ersity Gain in Urban Area Lo w Ele vation Links for Surveillance Applications, ” IEEE T rans. Antennas Propa g. , vol. 61, no. 12, pp. 6255-6260, 2013. [62] M. Simunek, P . Pechac and F .P . Fontan, ”Feasibility of UA V Link Space Diversity in W ooded Areas, ” Intl. J. Antennas Pr opag . , vol. 2013, pp. 1-6, 2013. [63] R.M. Gutierrez, H. Y u, Y . Rong and D.W . Bliss, ”Time and Frequency Dispersion Characteristics of the U AS W ire- less Channel in Residential and Mountainous Desert T errains, ” in Pr oc. IEEE Annual Consumer Commun. Netw . Conf. (CCNC’17) , Las V egas, USA, Jan. 2017, pp. 516-521. [64] H.T . Kung, C.K. Lin, T .H. Lin, S.J. T arsa and D. Vlah, ”Measuring Div ersity on a Low–Altitude U A V in a Ground–to–Air W ireless 802.11 Mesh Network, ” in Pr oc. IEEE Global Commun. Conf. (GLOBECOM’10) , Miami, USA, Dec. 2010, pp. 1799-1804. [65] Qualcomm T echnologies, Inc. ”L TE Unmanned Aircraft Systems Trial Report, ”2017, av ailable online at https://www .qualcomm.com/documents/lte-unmanned-aircraft-systems-trial-report. [66] Ericsson, ”Ericsson and China Mobile conduct worlds ﬁrst 5G drone prototype ﬁeld trial”, 2016, av ailable online at https://www .ericsson.com/en/ne ws/2016/8/ericsson-and-china-mobile-conduct-worlds-ﬁrst-5g-drone-prototype- ﬁeld-trial-. [67] L. Afonso, N. Souto, P . Sebastiao, M. Ribeiro, T . T av ares and R. Marinheiro, ”Cellular for the Skies: Exploiting Mobile Network Infrastructure for Low Altitude Air–to–Ground Communications, ” IEEE Aerosp. Electr on. Syst. , vol. 31, no. 8, pp. 4-11, 2016. [68] E. T eng, J. D. Falcao and B. Iannucci, ”Holes–in–the–Sky: A Field Study on Cellular -Connected U AS, ” in Proc. (ICU AS’17) , Miami, USA, June 2017, pp. 1165-1174. [69] S. Elnoubi, ”A Simpliﬁed Stochastic Model for the Aeronautical Mobile Radio Channel, ” in Pr oc. IEEE V ehicular T ech. Conf. (VTC’92) , Den ver , USA, May 1992, vol. 2, pp. 960-963. [70] M. A. Zaman, S. A. Mamun, M. Gaffar , M. M. Alam and M. I. Momtaz, ”Modeling VHF Air–to–Ground Multipath Propagation Channel and Analyzing Channel Characteristics and BER Performance, ” in Pr oc. IEEE Re gion 8 SIBIRCON , Listvyanka, Russia, July 2010, pp. 335-338. [71] T .J. W illink, C. Squires, G. Colman and M. Muccio, ”Measurement and Characterization of Low-Altitude Air-to-Ground MIMO Channels, ” IEEE T rans. V eh. T echnol. , v ol. 65, no. 4, pp. 2637-2648, 2016. [72] T .T . Georgiou, ”Distances and Riemannian metrics for spectral density functions, ” IEEE T rans. Signal Pr ocess ., vol. 55, no. 8, pp. 3995-4003, 2007. [73] T .J. W illink, ”Wide sense stationarity of mobile MIMO radio channels, ” IEEE Tr ans. V eh. T echnol. , vol. 57, no. 2, pp. 704-714, 2008. DRAFT January 24, 2018

A Survey of Channel Modeling for UAV Communications

Original Paper

Comments & Academic Discussion

Leave a Comment

Original Paper

Related Papers

Comments & Academic Discussion

Leave a Comment