3-D Statistical Indoor Channel Model for Millimeter-Wave and Sub-Terahertz Bands
Shihao Ju, Yunchou Xing, Ojas Kanhere, Theodore S. Rappaport
aa r X i v : . [ ee ss . SP ] S e p S. Ju, Y. Xing, O. Kanhere and T. S. Rappaport, “3-D Statistical Indoor Channel Model for Millimeter-Wave andSub-Terahertz Bands,” 2020
IEEE Global Communications Conference (GLOBECOM) , Dec. 2020, pp. 1-7.
Shihao Ju, Yunchou Xing, Ojas Kanhere, and Theodore S. Rappaport
NYU WIRELESS, NYU Tandon School of Engineering, Brooklyn, NY, 11201 { shao, ychou, ojask, tsr } @nyu.edu Abstract —Millimeter-wave (mmWave) and Terahertz(THz) will be used in the sixth-generation (6G) wirelesssystems, especially for indoor scenarios. This paper presentsan indoor three-dimensional (3-D) statistical channel modelfor mmWave and sub-THz frequencies, which is developedfrom extensive channel propagation measurements con-ducted in an office building at 28 GHz and 140 GHz in 2014and 2019. Over 15,000 power delay profiles (PDPs) wererecorded to study channel statistics such as the number oftime clusters, cluster delays, and cluster powers. All the pa-rameters required in the channel generation procedure arederived from empirical measurement data for 28 GHz and140 GHz line-of-sight (LOS) and non-line-of-sight (NLOS)scenarios. The channel model is validated by showing thatthe simulated root mean square (RMS) delay spread andRMS angular spread yield good agreements with measuredvalues. An indoor channel simulation software is built uponthe popular NYUSIM outdoor channel simulator, whichcan generate realistic channel impulse response, PDP, andpower angular spectrum.
Index Terms —Millimeter-Wave; Terahertz; Indoor Of-fice; Channel Measurement; Channel Modeling; ChannelSimulation; 5G; 6G
I. I
NTRODUCTION
Driven by ubiquitous usage of mobile devices andthe emergence of massive Internet of Things (IoT),the sixth-generation (6G) wireless system will offerunprecedented high data rate and system throughput.Moving to millimeter-wave (mmWave) and Terahertz(THz) frequencies (i.e., 30 GHz - 3 THz) is consideredas a promising solution of fulfilling future data trafficdemand created by arising wireless applications such asaugmented/virtual reality (AR/VR) and 8K ultra highdefinition (UHD) due to the vast bandwidth [1].Accurate THz channel characterization for indoor sce-narios facilitates the designs of transceivers, air interface,and protocols for 6G and beyond. Standards documentslike IEEE 802.11 ad/ay and 3GPP TR 38.901 proposedstatistical channel models for various indoor scenariossuch as office, home, shopping mall, and factory [2]–[4].The IEEE 802.11 ad/ay developed a double directionalstatistical channel impulse response (CIR) model work-ing at 60 GHz based on measurements and ray-tracingresults in the conference room, cubical environment, andliving room [2], [3]. 3GPP TR 38.901 presented a cluster-based statistical channel model for frequencies from 0.5 to 100 GHz for outdoor and indoor scenarios with differ-ent values of large-scale parameters such as delay spread,angular spread, Rician K-factor, and shadow fading [4].This paper proposes a unified indoor channel modelacross mmWave and sub-THz bands based on indoorchannel measurements at 28 GHz and 140 GHz, whichprovides a reference for future standards developmentabove 100 GHz.There has been some work on channel modeling atTHz bands [5]–[7]. A temporal-spatial statistical channelmodel based on ray-tracing results in an office room at275-325 GHz, which consisted of line-of-sight (LOS),first- and second-order reflected paths, was developed in[5]. A generic multi-ray channel model for 0.06-1 THzconstituted by LOS, reflected, diffracted, and scatteredpaths generated by ray-tracing simulations was given in[6]. Due to the hardware constraints such as maximumtransmit power, most of the published channel propa-gation measurements at THz frequencies were limitedwithin a few meters [7]. The existing channel mod-els for THz frequencies are mainly ray-tracing based,where CIRs are represented as a superposition of LOS,reflected, and scattered paths based on reflection andscattering properties of the environment. In this paper,the presented three-dimensional (3-D) channel model formmWave and sub-THz frequencies is statistical and builtupon extensive propagation measurements up to 45 m inan office building at 28 GHz and 140 GHz.The remainder of the paper is organized as follows.Section II describes the indoor measurements at 28 GHzand 140 GHz performed in 2014 and 2019. Section IIIintroduces the large-scale path loss model. Section IVdescribes the spatial statistical CIR model, and SectionV shows statistics of required parameters in the channelgeneration procedure. Section VI validates the proposedchannel model by showing that simulated and measuredomnidirectional root mean square (RMS) delay spread(DS) and global RMS angular spread (AS) yield a goodagreement. Section VII provides concluding remarks.II. W
IDEBAND C HANNEL M EASUREMENTS
A. Measurement environment, system, and procedure
28 GHz and 140 GHz channel measurement cam-paigns were conducted in the NYU WIRELESS re-earch center on the 9th floor of 2 MetroTech Centerin downtown Brooklyn, New York in 2014 and 2019,respectively. The 9th floor is a typical office environment(65.5 m ×
35 m × Fig. 1: Floor plan of the 9th floor, 2 MetroTech Center. Thereare five TX locations denoted as yellow stars and 33 RXlocations denoted as red dots [8].
We used a wideband sliding correlation-based channelsounding system with highly directional horn antennas torecord PDPs using a high-speed oscilloscope [9]. Thechannel sounder transmitted pseudorandom sequencesignals with 800 MHz and 1 GHz RF bandwidth at28 GHz and 140 GHz, respectively. Two identical hornantennas were used at TX and RX, which had 15 dBigain and 30° antenna half power beamwidth (HPBW)for 28 GHz measurements, and 27 dBi gain and 8° an-tenna HPBW for 140 GHz measurements. For each TX-RX location pair, eight unique antenna azimuth sweepswere performed to obtain spatial statistics for vertical-to-vertical (V-V) polarization. Overall, at most 96 (= × )at 28 GHz and 360 (= × ) at 140 GHz directionalPDPs were acquired with V-V polarization configurationfor each TX-RX location pair.For each antenna azimuth sweep, the RX (TX) hornantenna was swept in steps of 8° or 30° accordingto antenna HPBW, and the pointing direction of TX(RX) horn antenna was fixed during one sweep. At eachsweeping step, an average PDP over 20 instantaneousPDPs was recorded by a high-speed oscilloscope for next-step processing and analysis. By sweeping anten-nas in the azimuth plane at different elevation levels,multipath components (MPCs) with energy above thenoise floor that can arrive at RX in the 3-D space werecaptured and recorded. More details of the measurementprocedure can be found in [8], [10]. B. Data processing
Omnidirectional channel modeling is fundamental andregarded as a basis for directional and multiple inputmultiple output (MIMO) channel modeling [11]. Thus,omnidirectional PDPs need to be recovered from mea-sured directional PDPs with knowledge of absolute timedelays. Note that the channel sounding system was notable to perform precise synchronization between TX andRX, thus cannot provide absolute timing informationbecause the PDP recording was triggered at the firstMPC arrival and only had excess time delay information.Here we used a ray-tracing tool to provide the actualtime of flight of the first MPC in measured directionalPDPs so that an omnidirectional PDP can be synthesizedby aligning directional PDPs in time at each TX-RXlocation pair.A 3-D mmWave ray-tracing tool, NYURay [12], wasused to predict possible propagating rays for the identicalTX-RX location pairs selected in measurements, whichcan provide time of flight of the predicted rays. If thedirection of a measured directional PDP matched thedirection of a predicted ray, the time of flight of thispredicted ray was used as the absolute time delay of thefirst MPC in this measured PDP. Due to the beamwidthof horn antennas, there might be several predicted rayswhich were close to the pointing angle of the measuredPDP in space and vice versa. The predicted ray whichwas closest to the measured PDP in 3-D space (i.e.,the minimal sum of the differences of azimuth angleof departure (AOD) φ AOD , zenith angle of departure(ZOD) θ ZOD , azimuth angle of arrival (AOA) φ AOA ,and zenith angle of arrival (ZOA) θ ZOA ) was chosento match this measured PDP and provide the absolutetiming information.III. L
ARGE - SCALE P ATH L OSS M ODEL
Path loss models describe the distance-power law thatthe received power decreases exponentially with distanceand are commonly used in the prediction of signalstrength and cell range. A popular path loss model, close-in free space reference distance (CI) path loss model with1 m reference distance, is given by [8], [13], [14]PL CI ( f, d )[ dB ] = FSPL ( f, m )[ dB ] + 10 n log ( d ) + χ σ , (1)where n is path loss exponent (PLE) and d is thedistance. FSPL ( f, m ) is the free space path loss at 1m at frequency f . Shadow fading χ σ is characterizedy a zero mean Gaussian random variable with standarddeviation σ in dB.PLE n indicates that the power decays by 10 n dBper decade of distance beyond 1 m [14]. Fig. 2 showsthat PLE for 28 GHz LOS and NLOS scenarios viaminimum mean square error (MMSE) fitting are 1.2 and2.8, respectively. 1.2 is derived from 2 MetroTech datasetfor 28 GHz LOS case, which is lower than 1.3-1.9 foundin the literature [15]–[17]. To verify this low PLE, weconducted LOS measurements in another office building,370 Jay, at 28 GHz. The resulting PLE for 370 Jaydataset is also 1.2, suggesting that the power attenuatesmuch slower (12 dB per ten meters) than values reportedin the literature, which might be attributed to the strongwaveguide effect of long and narrow corridors in theindoor environment. Fig. 2: 28 GHz indoor omnidirectional path loss scatter plotand MMSE-fitted CI path loss model with distance for LOSand NLOS scenarios.
IV. S
TATISTICAL O MNIDIRECTIONAL C HANNEL I MPULSE R ESPONSE M ODEL
Here we presented a 3GPP-like spatial statisticalchannel model for indoor scenarios using the NYUSIMoutdoor channel modeling framework [11]. The outdoorand indoor NYUSIM channel model share the channelgeneration procedure and the set of channel parameters,but have different probabilistic distributions for thesechannel parameters such as the number of time clusters(TCs) and cluster subpaths (SPs) due to the distinctenvironment characteristics. The proposed indoor chan-nel model is built upon the existing NYUSIM outdoormodeling procedure and developed from 28 GHz and140 GHz measurements in the office building.The NYUSIM and 3GPP TR 38.901 channel modelsare both 3-D statistical channel models but have severalkey differences [4], [18]. The 3GPP channel modeldefines a cluster as a group of MPCs closely spacedjointly in the temporal and spatial domain [4] whilethe NYUSIM channel model separately characterizes temporal and spatial statistics by defining time clusterand spatial lobe (SL) motivated by the observations frommeasurements [8]. A TC is composed of MPCs travelingclose in time, and that arrive from potentially differentdirections in a short propagation time window. An SLrepresents a main direction of arrival or departure whereMPCs may arrive over hundreds of ns [11].Both modeling methodologies are valid, where the3GPP model is more widely used and the NYUSIMmodel has a simpler and physically-based structure [19],[20]. Performance evaluation with respect to spectrumefficiency, coverage, and hardware/signal processing re-quirements between the 3GPP and NYUSIM channelmodels were analyzed in detail [18].A received signal can be regarded as a superposition ofmultiple replicas of the transmitted signal with differentdelays and angles for any wireless propagation channel.The time cluster spatial lobe (TCSL)-based omnidirec-tional CIR model is given by h omni ( t, Θ , Φ) = N X n =1 M n X m =1 a m,n e jϕ m,n · δ ( t − τ m,n ) · δ ( −→ Θ − −−−→ Θ m,n ) · δ ( −→ Φ − −−−→ Φ m,n ) , (2)where t is the absolute propagation time, −→ Θ =( φ AOD , θ
ZOD ) is the vector of AOD and ZOD, and −→ Φ = ( φ AOA , θ
ZOA ) is the vector of AOA and ZOA. N and M n denote the number of TCs and the numberof cluster SPs, respectively. For the m th SP in the n thTC, a m,n , ϕ m,n , τ m,n , −−−→ Θ m,n , and −−−→ Φ m,n represent themagnitude, phase, absolute propagation time, AOD andAOA, respectively. The PDP and power angular spectrum(PAS) can be obtained by integrating the square of theCIR in space and time domains, respectively.The measured PDP and PAS were partitioned into timeclusters and spatial lobes to obtain empirical channelstatistics such as the number of TCs and the numberof SLs, respectively [11]. The partition in the temporaldomain was realized by minimum inter-cluster time voidinterval (MTI). Two sequentially recorded SPs belongto two distinct time clusters if the difference of theexcess time delays of these two SPs is beyond MTI. Forexample, 25 ns was used as MTI for an outdoor urbanmicrocell (UMi) environment [11], while 6 ns was usedas MTI in this paper for an indoor office environmentdue to the fact that the width of a typical hallway inthe measured indoor office environment was about 1.8m (i.e., ∼ TATISTICS E XTRACTION OF C HANNEL P ARAMETERS
Temporal and spatial channel parameters are extractedfrom the measured PDP and PAS using the TCSLapproach described in Section IV. Temporal parametersinclude the number of TCs ( N ) and SPs in a TC ( M n ),TC excess delay ( τ n ) and intra-cluster SP excess delay( ρ m,n ), TC power ( P n ) and SP power ( Π m,n ). Spatialparameters are the number of SLs ( L ), the mean azimuthand elevation angle of an SL ( φ and θ ) and the azimuthand elevation angular offset of a SP ( ∆ φ and ∆ θ ) withrespect to the mean angle of the SL. Values of parametersrequired in Table I for 28 GHz and 140 GHz LOS andNLOS scenarios are given in Table II, where DU standsfor discrete uniform. A. Temporal Channel Parameters1) The number of time clusters:
The empirical his-togram of the number of TCs N from 28 GHz and 140GHz NLOS measurements with a 6 ns MTI is shownin Fig. 3, which follows a Poisson distribution. SincePoisson distribution starts from zero while the numberof TCs is at least one, N ′ = N − is used for distributionfitting. The generated number of TCs from the Poissondistribution is added by one to obtain the simulatednumber of TCs, which is given by P ( N ′ = k ) = λ kc k ! e − λ c , k = 0 , , , ...,N = N ′ + 1 . (3)where λ c is the mean of the Poisson distribution. Fig. 3shows that there are more time clusters at 28 GHz than140 GHz which is likely due to the higher partition lossat 140 GHz (e.g., 4-8 dB higher than 28 GHz) [10]. ThePoisson distribution of the number of TCs for the indoorNLOS scenario is different from the uniform distributionfor the outdoor scenario [11].
2) Number of cluster subpaths:
The number of clusterSPs M n is related to the number of TCs and dependson the choice of MTI. A larger MTI causes fewer TCsand more SPs per TC since two SPs might be countedinto one cluster rather than two clusters using a largerMTI. The empirical histogram of the number of clustersubpaths from 28 GHz measurements with a 6 ns MTI isshown in Fig. 4. Similar to the number of time clusters, M ′ n = M n − was used for distribution fitting. Weproposed a composite distribution with a δ -function at Fig. 3: The number of time clusters for 28 GHz and 140 GHzNLOS scenario follows Poisson distributions with mean 3.4and 1.3, respectively.Fig. 4: The number of cluster subpaths for 28 GHz LOS andNLOS scenarios follows the composite distribution. M ′ n = 0 and a discrete exponential (DE) distribution,which is given by P M ′ n ( k ) = (1 − β ) δ ( k ) + β Z k +1 k µ s e − xµs dx,k = 0 , , , ..., (4)where µ s is the mean of the DE distribution and β is theweight of the DE distribution in the composite distribu-tion. The maximum likelihood estimation (MLE) of µ s and β are 4.1 and 0.6 for 28 GHz NLOS measurements,respectively. The identical composite distribution for the28 GHz LOS scenario shows that µ s = 2 . and β = 0 . ,suggesting that NLOS scenario forms relatively largerclusters than the LOS scenario. As shown in Table II, interms of the number of TCs (depending on the inputparameter λ c ) and the number of SPs (depending onthe input parameter µ s ), the 140 GHz channel is much ABLE I: I
NPUT P ARAMETERS FOR CHANNEL COEFFICIENT GENERATION PROCEDURE
Step Index Channel Parameters 28 - 140 GHz InHStep 1
N N ∼ ( DU (1 , N c ) , if LOSPoisson ( λ c ) , if NLOS Step 2 M n M n ∼ (1 − β ) δ ( M n ) + β · DE ( µ s ) Step 3 Intra-cluster delay ρ m,n (ns) ρ m,n ∼ Exp ( µ ρ ) Step 4 Cluster delay τ n (ns) τ ′ n ∼ Exp ( µ τ ) or Logn ( µ τ , σ τ )∆ τ n = sort ( τ ′ n ) − min ( τ ′ n ) τ n = ( , n = 1 τ n − + ρ M n − ,n − + ∆ τ n + MTI , n = 2 , ...N
Step 5 Cluster power P n (mW) P ′ n = ¯ P e − τn Γ Zn ,P n = P ′ n P Nk =1 P ′ k × P r [ mW ] ,Z n ∼ N (0 , σ Z ) , n = 1 , , ..., N Step 6 Subpath power Π m,n (mW) Π ′ m,n = ¯Π e − ρm,nγ Um,n , Π m,n = Π ′ m,n P M n k =1 Π ′ k,n × P n [ mW ] ,U m,n ∼ N (0 , σ U ) , m = 1 , , ..., M n Step 7 SP phase ϕ (rad) Uniform(0, 2 π ) Step 8
L L
AOD ∼ DU (1 , L AOD,max ) L AOA ∼ DU (1 , L AOA,max ) Step 9 SL mean angle φ i , θ i (°) φ i ∼ Uniform ( φ min , φ max ) ,φ min = i − L , φ max = iL , i = 1 , , ..., Lθ i ∼ N ( µ l , σ l ) Step 10 SP angle offset ∆ φ i , ∆ θ i w.r.t φ i , θ i (°) i ∼ DU [1 , L AOD ] , j ∼ DU [1 , L AOA ](∆ φ i ) m,n, AOD ∼ N (0 , σ φ, AOD )(∆ θ i ) m,n, ZOD ∼ N (0 , σ θ, ZOD )(∆ φ j ) m,n, AOA ∼ N (0 , σ φ, AOA )(∆ θ j ) m,n, ZOA ∼ N (0 , σ θ, ZOA ) sparser than the 28 GHz channel, which is critical forchannel estimation [21].
3) Inter- and intra-cluster excess delay:
The clusterexcess delay τ n is defined as the time difference betweenthe first arriving subpath in the PDP and the first arrivingsubpath in the n th cluster, as shown in Step 4 of TableI. ρ M n − ,n − is the intra-cluster excess delay of the lastsubpath in the former cluster. ∆ τ n is the inter-clusterexcess delay without MTI (i.e., 6 ns). Results given inTable II show that the inter-cluster delays for 28 GHzNLOS, 140 GHz LOS, and 140 GHz NLOS datasets arewell characterized by exponential distributions. However,the inter-cluster delay for 28 GHz LOS dataset is closerto a lognormal distribution. The peculiar behavior foundin 28 GHz LOS scenario is attributed to two strong MPCswith large inter-cluster delays observed in the corridorenvironment (e.g., TX4 and RX 16).The intra-cluster excess delay ρ m,n is defined as thetime difference between the first arriving subpath andthe m th arriving subpath in the n th time cluster. Anexponential distribution with µ ρ shows a good agreementwith the measured intra-cluster excess delay for 28 GHz and 140 GHz LOS and NLOS scenarios. B. Spatial Channel Parameters1) The number of spatial lobes:
An SL representsa main direction of arrival or departure. The angularresolution of the measured PAS depends on the antennaHPBW (30° and 8° for 28 and 140 GHz measurements).Linear interpolation of the measured PAS with 1° angularresolution in the azimuth and elevation planes was imple-mented to investigate the 3-D spatial distribution of thereceived power. Measurement results showed that thereare at most two main directions of arrival in the azimuthplane, except that there are a few NLOS locations havingthree main directions of arrival in 28 GHz measurements.
2) Spatial lobe mean angle and subpath angular off-set:
The SL mean azimuth angle φ i is generated by firstdividing the azimuth plane into a few sectors accordingto the number of spatial lobes, then randomly selectinga direction within that sector. The SL mean elevationangle θ i is generated by a normal random variable withthe mean µ l . The angular offsets of each subpath in bothzimuth and elevation planes are modeled as zero-meannormal random variables with variances σ φ and σ θ .VI. S IMULATION R ESULTS
The statistical channel model presented in SectionV was implemented as an indoor channel simulatorbased on the NYUSIM outdoor channel simulator toinvestigate the accuracy of the simulated temporal andspatial statistics by comparing it with the measuredstatistics. Note that the parameters listed in Table IIare primary statistics which are used in the channelparameter generation procedure. The metrics used in thissection for channel validation are secondary statisticssuch as RMS DS and RMS AS which are not explic-itly used in the channel generation, but the simulatedsecondary statistics should yield good agreements withthe measured statistics. 10,000 simulations were carriedout for four frequency scenarios (i.e., 28 GHz LOS, 28GHz NLOS, 140 GHz LOS, and 140 GHz NLOS) bygenerating 10,000 omnidirectional PDPs, and 3-D AODand AOA PASs as sample functions of (2) using theNYUSIM indoor channel simulator.
A. Simulated RMS Delay Spreads
The RMS DS describes channel temporal dispersion,which is a critical metric to validate a statistical channelmodel. Fig. 5 shows the simulated and measured omnidi-rectional RMS DS at 28 GHz and 140 GHz in LOS andNLOS scenarios. As shown in Fig. 5, the empirical andsimulated median RMS DS are 17.9 and 13.9 ns for 28GHz LOS scenario, 13.5 and 12.5 ns for 28 GHz NLOSscenario, 3.1 and 3.2 ns for 140 GHz LOS scenario, and5.7 and 5.9 ns for 140 GHz NLOS scenario, respectively.The simulated cumulative distribution function (CDF)yielded good agreements with the empirical CDF for fourfrequency scenarios.
Fig. 5: Measured and simulated omnidirectional RMS DS for28 and 140 GHz in LOS and NLOS scenarios. Fig. 6: Measured and simulated global RMS AOA AS for 28and 140 GHz in LOS and NLOS scenarios.
B. Simulated RMS Angular Spreads
The omnidirectional azimuth and elevation AS de-scribe the angular dispersion at a TX or RX over theentire 4 π steradian sphere, also termed global AS. TheAOA and AOD global ASs were computed using the total(integrated over delay) received power over all measuredazimuth/elevation pointing angles. The measured andsimulated global AOA RMS AS was calculated usingAppendix A-1,2 in [4] and compared in Fig. 6, show-ing the simulated and measured median global angularspreads are very close (less than 5°) for 28 GHz LOS, 28GHz NLOS, and 140 GHz NLOS cases while the limitednumber of 140 GHz LOS measurements considerablyskewed the empirical distribution.VII. C ONCLUSION
The paper presented a 3-D spatial statistical channelmodel based on the extensive measurements at 28 and140 GHz in an indoor office building. The statistics ofnecessary parameters for the channel generation pro-cedure were extracted from empirical LOS and NLOSmeasurement data. The NYUSIM indoor channel simu-lator was used to generate thousands of PDPs and PASsfor validating the presented channel model. The simu-lated channel statistics yielded a good agreement withthe measured channel statistics in terms of secondarystatistics such as omnidirectional RMS DS and globalRMS AS. The omnidirectional statistical channel modelin this work will be used as a basis for further directionaland MIMO channel modeling and facilitates channelestimation and channel capacity analysis for mmWaveand sub-THz frequencies.
ABLE II: R
EQUIRED PARAMETERS THAT REPRODUCE THE MEASURED STATISTICS OF OMNIDIRECTIONAL CHANNELS USINGTHE PRESENTED STATISTICAL CHANNEL MODEL . Input Parameters 28 GHz LOS 28 GHz NLOS 140 GHz LOS 140 GHz NLOS N c λ c NA 3.4 NA 1.3 β s µ s µ τ [ ns ] logn(2.7, 1.4) 12.1 18.6 23.5 µ ρ [ ns ] Γ[ ns ] , σ Z [ dB ] γ [ ns ] , σ U [ dB ] L AOD,max , L
AOA,max
2, 2 2, 3 2, 2 2, 2 µ l, ZOD [ ° ] , σ l, ZOD [ ° ] -7.3, 3.8 -5.5, 2.9 -6.8, 4.9 -2.5, 2.7 µ l, ZOA [ ° ] , σ l, ZOA [ ° ] σ φ, AOD [ ° ] , σ θ, AOD [ ° ] σ φ, AOA [ ° ] , σ θ, AOA [ ° ] R EFERENCES[1] T. S. Rappaport et al. , “Wireless communications and applica-tions above 100 GHz: Opportunities and challenges for 6G andbeyond,”
IEEE Access , vol. 7, pp. 78 729–78 757, June 2019.[2] A. Maltsev et al. , “Channel models for 60 GHz WLAN systems,”doc.: IEEE 802.11-09/0334r8, May 2010.[3] ——, “Channel models for IEEE 802.11ay,” doc.: IEEE 802.11-15/1150r9, May 2016.[4] 3GPP, “Technical specification group radio access network; studyon channel model for frequencies from 0.5 to 100 GHz (Release16),” TR 38.901 V16.0.0, October 2019.[5] S. Priebe and T. Kurner, “Stochastic modeling of THz indoorradio channels,”
IEEE Transactions on Wireless Communications ,vol. 12, no. 9, pp. 4445–4455, September 2013.[6] C. Han, A. O. Bicen, and I. F. Akyildiz, “Multi-ray channelmodeling and wideband characterization for wireless communi-cations in the Terahertz band,”
IEEE Transactions on WirelessCommunications , vol. 14, no. 5, pp. 2402–2412, May 2015.[7] C. Cheng, S. Sangodoyin, and A. Zaji, “THz cluster-basedmodeling and propagation characterization in a data center envi-ronment,”
IEEE Access , vol. 8, pp. 56 544–56 558, March 2020.[8] G. R. Maccartney, T. S. Rappaport, S. Sun, and S. Deng, “Indooroffice wideband millimeter-wave propagation measurements andchannel models at 28 and 73 GHz for ultra-dense 5G wirelessnetworks,”
IEEE Access , vol. 3, pp. 2388–2424, October 2015.[9] G. R. MacCartney and T. S. Rappaport, “A flexible millimeter-wave channel sounder with absolute timing,”
IEEE Journal onSelected Areas in Communications , vol. 35, no. 6, pp. 1402–1418,March 2017.[10] Y. Xing, T. S. Rappaport, and A. Ghosh, “Millimeter wave andsub-Terahertz indoor office building propagation measurementsand channel models at 28, 73, and 142 GHz,” submitted to IEEEJournal on Selected Areas in Communications , April 2020.[11] M. K. Samimi and T. S. Rappaport, “3-D millimeter-wave sta-tistical channel model for 5G wireless system design,”
IEEETransactions on Microwave Theory and Techniques , vol. 64,no. 7, pp. 2207–2225, July 2016.[12] O. Kanhere et al. , “Map-assisted millimeter wave localizationfor accurate position location,” in , December 2019, pp. 1–6.[13] S. Sun et al. , “Investigation of prediction accuracy, sensitivity, andparameter stability of large-scale propagation path loss models for5G wireless communications,”
IEEE Transactions on VehicularTechnology , vol. 65, no. 5, pp. 2843–2860, May 2016. [14] T. S. Rappaport et al. , “Wideband millimeter-wave propagationmeasurements and channel models for future wireless commu-nication system design (Invited Paper),”
IEEE Transactions onCommunications , vol. 63, no. 9, pp. 3029–3056, September 2015.[15] J. Ko, S. U. Lee, Y. S. Kim, and D.-J. Park, “Measurementsand analyses of 28 GHz indoor channel propagation based ona synchronized channel sounder using directional antennas,”
Journal of Electromagnetic Waves and Applications , vol. 30,no. 15, pp. 2039–2054, June 2016.[16] Hao Xu, V. Kukshya, and T. S. Rappaport, “Spatial and temporalcharacteristics of 60-GHz indoor channels,”
IEEE Journal onSelected Areas in Communications , vol. 20, no. 3, pp. 620–630,April 2002.[17] L. Rubio et al. , “Wideband propagation channel measurements inan indoor office environment at 26 GHz,” in , October 2019, pp. 2075–2076.[18] S. Sun, T. S. Rappaport, M. Shafi, P. Tang, J. Zhang, andP. J. Smith, “Propagation models and performance evaluationfor 5G millimeter-wave bands,”
IEEE Transactions on VehicularTechnology , vol. 67, no. 9, pp. 8422–8439, June 2018.[19] M. Polese, M. Giordani, M. Mezzavilla, S. Rangan, and M. Zorzi,“Improved handover through dual connectivity in 5G mmwavemobile networks,”
IEEE Journal on Selected Areas in Communi-cations , vol. 35, no. 9, pp. 2069–2084, June 2017.[20] C. N. Barati, S. A. Hosseini, S. Rangan, P. Liu, T. Korakis,S. S. Panwar, and T. S. Rappaport, “Directional cell discoveryin millimeter wave cellular networks,”
IEEE Transactions onWireless Communications , vol. 14, no. 12, pp. 6664–6678, July2015.[21] C. Huang, L. Liu, C. Yuen, and S. Sun, “Iterative channelestimation using lse and sparse message passing for mmwavemimo systems,”