Portal de Periódicos da CAPES

Measurement-Based Propagation Channel Characteristics for Millimeter-Wave 5G Giga Communication Systems

2016; Electronics and Telecommunications Research Institute; Volume: 38; Issue: 6 Linguagem: Inglês

10.4218/etrij.16.2716.0050

2233-7326

Juyul Lee, Jinyi Liang, Myung-Don Kim, Jae-Joon Park, Bonghyuk Park, Hyun Kyu Chung,

Advanced MIMO Systems Optimization

ETRI JournalVolume 38, Issue 6 p. 1031-1041 ArticleFree Access Measurement-Based Propagation Channel Characteristics for Millimeter-Wave 5G Giga Communication Systems Juyul Lee, Corresponding Author Juyul Lee juyul@etri.re.kr Corresponding Authorjuyul@etri.re.krSearch for more papers by this authorJinyi Liang, Jinyi Liang liangjinyi@etri.re.kr Search for more papers by this authorMyung-Don Kim, Myung-Don Kim mdkim@etri.re.kr Search for more papers by this authorJae-Joon Park, Jae-Joon Park jjpark@etri.re.kr Search for more papers by this authorBonghyuk Park, Bonghyuk Park bhpark@etri.re.kr Search for more papers by this authorHyun Kyu Chung, Hyun Kyu Chung hkchung@etri.re.kr Search for more papers by this author Juyul Lee, Corresponding Author Juyul Lee juyul@etri.re.kr Corresponding Authorjuyul@etri.re.krSearch for more papers by this authorJinyi Liang, Jinyi Liang liangjinyi@etri.re.kr Search for more papers by this authorMyung-Don Kim, Myung-Don Kim mdkim@etri.re.kr Search for more papers by this authorJae-Joon Park, Jae-Joon Park jjpark@etri.re.kr Search for more papers by this authorBonghyuk Park, Bonghyuk Park bhpark@etri.re.kr Search for more papers by this authorHyun Kyu Chung, Hyun Kyu Chung hkchung@etri.re.kr Search for more papers by this author First published: 01 December 2016 https://doi.org/10.4218/etrij.16.2716.0050Citations: 44 Juyul Lee (corresponding author, juyul@etri.re.kr), Jinyi Liang (liangjinyi@etri.re.kr), Myung-Don Kim (mdkim@etri.re.kr), Jae-Joon Park (jjpark@etri.re.kr), Bonghyuk Park (bhpark@etri.re.kr), and Hyun Kyu Chung (hkchung@etri.re.kr) are with the 5G Giga Communication Research Laboratory, ETRI, Daejeon, Rep. of Korea This work was supported by the Institute for Information & Communications Technology Promotion (IITP) grant funded by the Korean government (MSIP) (B0101-16-222, Development of core technologies to improve spectral efficiency for mobile big-bang) and (B0194-15-1002, Korea-China bilateral and international collaboration of mmWave core technologies for 5G standardization). When specifying 5G mmWave frequencies, it is customary to refer to the frequencies between 10 GHz and 100 GHz [2]. The original SAGE is designed to process array antenna measurements [15]. The link type (LOS or NLOS) is determined by the existence of an unobstructed straight-line path between the TX and RX locations regardless of antenna types. This is to say that the boresights of TX and RX antennas are not necessarily coaxial for the LOS conditions when directional antennas are used. We were able to obtain a large number of omnidirectional antenna measurement points (compared with 10° HPBW antenna measurements) because we collected additional measurement points while moving the RX equipment, instead of rotating the antenna at a single fixed location. We used two antennas for each of the frequency bands. Because HPBW and antenna gain vary with operating frequency, we prepared two different sets of antennas for 28 GHz and 38 GHz measurements. We consider only a single slope model with respect to distance, as given in (1) and (2), because the break point distance is significantly greater than the expected cell radius of 200 m. For our measurement setup at 28 GHz and 38 GHz, the break point distances (hTXhRXf/c for urban scenarios [14]) are 5,600 m and 7,600 m, respectively. AboutFiguresReferencesRelatedInformationPDFSectionsAbstract I. Introduction II. Measurement Overview III. Path Loss Model IV. Multipath Characteristics V. ConclusionBiographiesReferencesCiting LiteraturePDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessClose modalShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Abstract This paper presents millimeter-wave (mmWave) propagation characteristics and channel model parameters including path loss, delay, and angular properties based on 28 GHz and 38 GHz field measurement data. We conducted measurement campaigns in both outdoor and indoor at the best potential hotspots. In particular, the model parameters are compared to sub-6 GHz parameters, and system design issues are considered for mmWave 5G Giga communications. For path loss modeling, we derived parameters for both the close-in free space model and the alpha–beta–gamma model. For multipath models, we extracted delay and angular dispersion characteristics including clustering results. I. Introduction In order to provide hyperconnected mobile networks with gigabyte contents, the Korean government initiated a megascale project in 2012, referred to as Giga Korea(GK) [1]. As a government research institute, ETRI has been developing millimeter-wave (mmWave)1) stand-alone 5G cellular networks (titled GK-5G) since 2013 [3]. To facilitate successful system development in mmWave frequency bands that have yet to be explored for mobile cellular networks, the understanding of mmWave propagation characteristics is necessary not only for system design but also for verification and evaluation. In this study, we investigate mmWave propagation characteristics based on field measurement data used in the mmWave-based 5G cellular system of the GK-5G project. The field measurement data were collected in an urban microcellular (UMi) environment and two indoor hotspot (InH) areas at 28 GHz and 38 GHz: a downtownn area in Daejeon city was selected (for UMi) and the passenger terminals of Seoul Railway Station and Incheon International Airport were selected (for InH). In contrast to frequency bands below 6 GHz, the use of mmWave frequencies for mobile cellular networks has been widely debated because of misconceptions on mmWave propagation behavior with respect to path loss and straight-line-of-sight propagation properties. These myths were commonly accepted because of their apparent inherency to very high frequencies. In this study, we consider these issues by investigating mmWave propagation characteristics and channel models, specifically targeting mobile cellular networks. Recently, numerous channel measurement and modeling results have been reported by academia, research institutes, and telecommunication companies to promote mmWave channel model standards for 5G [4]–[11]. By rotating narrow-beamwidth horn antennas, researchers at New York University conducted 28 GHz measurements and reported path loss and multipath angular and delay characteristics [4] and claimed that "mmWave will work for 5G" considering that most hotspot cell sizes will be less than 200 m. They also reported omnidirectional characteristics from directional-antenna measurements considering the importance of omnidirectional characteristics for system design and evaluation [5], [6]. It should also be noted that most standardized channel models such as 3GPP [12], [13] and ITU-R [14] are provided with omnidirectional antennas. Numerous synthesis methods exist, and each method produces slightly different characteristics. In this respect, we conducted measurements with directional horn antennas as well as omnidirectional antennas. Omnidirectional antennas were specifically employed vice the more common scanned directional antenna technique [4]–[6]. The intention is to avoid errors that might normally occur when synthesis algorithms and methodologies are used. Our approach involves fitting of our 28 GHz and 38 GHz measurement data to existing 3GPP/ITU-R channel models that have been successfully employed for 3G and 4G systems and were proven to be valid for frequencies below 6 GHz. We then compare the mmWave parameters extracted from our measurements with those of sub-6 GHz models and discuss the impacts on mmWave system designs. We use omnidirectional antenna measurements for large-scale propagation behaviors such as path loss, and directional antenna measurements for multipath component level analysis. In order to extract multipath components from measurements, we employed a modified space-alternating generalized expectation-maximization (SAGE) algorithm2) that can process directionally scanned data [16], instead of examining the power delay profiles obtained by power measurements [4], [17]. By analyzing the multipath components, we developed a geometry-based stochastic spatial channel model. In order to achieve consensus prior to any attempt at standardization, the METIS project was conducted in Europe, and its mmWave channel models were released [18]. The METIS models, which were developed with preliminary measurement data and ray tracing, consist of a stochastic model, a map-based model, and a hybrid of these two. In EU, another project referred to as MiWEBA was conducted in collaboration with Japan to develop outdoor 60 GHz channel models [19]. A white paper on mmWave channel models was released at the IEEE Globecom 2015 conference and updated in March 2016 [20]. We contributed our UMi and InH measurement results to the white paper. Compared to the white paper contribution, this paper presents measurements and in-depth analysis including comparisons with sub-6GHz channel models that were not presented in the white paper. Furthermore, we will also discuss system design considerations with the obtained model parameters. Some other parts of the white paper contributed by other parties can be also found in [17], [21], [22]. In order to investigate mmWave wideband channel characteristics, we developed a channel sounder operating at 28 GHz and 38 GHz. It should be noted that these frequency bands are included in the mmWave candidate bands for 5G at the World Radiocommunication Conference 2015 [23]. Moreover, the 5G system in the GK-5G project is being developed in the 28 GHz frequency band [3]. Our 28 GHz and 38 GHz measurements will facilitate the development of 5G systems in those bands. II. Measurement Overview In order to conduct outdoor and indoor field measurement campaigns, we designed and built a wideband channel sounder operating at 28 GHz and 38 GHz. This sounder was developed using the swept time-delay cross-correlation method with a sliding correlation technology [24]. Figure 1 shows the channel sounder, and Table 1 lists the sounder specifications. Table 1. Channel sounder specifications. System parameter Specification 28 GHz 38 GHz Carrier center frequency 28.0 GHz 38.0 GHz Channel bandwidth 500 MHz 500 MHz PN code length 4,095 chips 4,095 chips Sliding factor 12,500 12,500 Maximum TX power 29 dBm 21 dBm AGC range 60 dB 60 dB Multipath delay resolution 2 ns 2 ns 10° HPBW horn antenna gain 24.4 dBi 24.6 dBi 30° HPBW horn antenna gain 15.4 dBi 16.4 dBi 40° HPBW horn antenna gain N/A 12.6 dBi 60° HPBW horn antenna gain 9.9 dBi N/A Omnidirectional antenna gain 5 dBi 6 dBi As shown in Fig. 1, the sounder is composed of a baseband module (BBM), a transceiver module (TRXM), a timing module (TIM), a 28/38 GHz RF module (RFM), and antennas. The RFM and one antenna are installed on the 3D positioner (shown in the figure) to vary the boresight direction of antenna in 1° steps. In the TIM, rubidium oscillators are installed to provide time synchronization between the TX and RX. Figure 1Open in figure viewerPowerPoint mmWave wideband channel exploration sounder system. 1. Measurement Scenarios By targeting 5G mmWave channel model developments, we selected the best candidate outdoor and indoor scenarios: outdoor UMi and indoor InH. In each environment, we collected measurements from at least two different TX locations. The measurements of each environment are given below. A. Outdoor Urban Microcellular (UMi) Environment For collecting UMi environment measurement data, we selected Dunsan district in Daejeon, South Korea, shown in Fig. 2. The measurement site is a typical downtown area composed of rectangular street grids. The average building height is 20 m–35 m, and the average street width is 24 m–35 m. As shown in Fig. 2, we selected two TX locations. Depending on the LOS and NLOS conditions,3) we selected RX locations, as shown in the figure. The number of measurement points4) and distance information are shown in Table 2. Table 2. Number of TX–RX measurement points and distance. Link type Freq. 10° HPBW ant. Omnidir ant. No. pts. Distance No. pts. Distance UMi LOS 28 GHz 24 55 m–171 m 253 55 m–171 m 38 GHz 24 55 m–171 m 344 55 m–171 m NLOS 28 GHz 46 43 m–131 m 284 43 m–171 m 38 GHz 47 43 m–131 m 309 43 m–131 m InH LOS 28 GHz 11 27 m–133 m 281 22 m–127 m 38 GHz 20 22 m–127 m 393 22 m–127 m NLOS 28 GHz 16 41 m–307 m 392 41 m–205 m 38 GHz 21 24 m–127 m 334 41 m–205 m At the TX side, we installed 30° half-power beamwidth (HPBW) antennas for both 28 GHz and 38 GHz5) measurements and positioned the boresight so as to cover the entire range of interest (shown as yellow shaded regions in Fig. 2) to emulate an omnidirectional antenna installation. As typical UMi scenarios suggested in [14], we installed the TX antenna at a height of 10 m, as shown in Fig. 3. At the RX side, we installed a 10° HPBW directional antenna and an omnidirectional antenna for both 28 GHz and 38 GHz measurements, as shown in Fig. 3. The directional antenna was rotated in 10° steps to capture multipath components from all the directions. In order to emulate pedestrian-use conditions, both the RX antennas were installed at 1.5 m. The effects of antenna gain/patterns, cable loss, and system impairments were removed by conducting a back-to-back calibration process [25]. Figure 2Open in figure viewerPowerPoint UMi measurement layout. Figure 3Open in figure viewerPowerPoint UMi measurement campaign. B. Indoor Hotspot (InH) Environment We conducted InH measurement campaigns at the passenger terminals of Seoul Railway Station and Incheon International Airport. These two measurement sites are relatively large and are the most busy railway station and airport terminal in Korea. The mobile data traffic demands in these sites are expected to be significant. The layout of each site is shown in Fig. 4, and the number of measurement points and distances are listed in Table 2. Figure 4Open in figure viewerPowerPoint InH measurement layout: (a) railway and (b) airport site. The sizes of the halls are approximately (height) and (height) in the railway and airport sites, respectively. The ceilings and walls of both the sites are made with steel frames and thick tempered glasses. Large LCD screens displaying train or flight departure/arrival information are mounted on the wall. One difference between the two sites is the arrangement of check-in booths in the airport site that are a major source of NLOS obstruction. Similarly, for UMi, we installed a relatively wide beamwidth antenna at the TX (60° HPBW antenna at 28 GHz and 40° at 38 GHz) to mimic an omnidirectional antenna installation by covering the entire range of interest. At the RX, both narrow-beamwidth directional (10° HPBW) and omnidirectional antennas were installed. When conducting directional antenna measurements, we rotated the antenna in 10° steps. The TX and RX antennas were installed at 8 m (ceiling level) and 1.5 m (pedestrian level). III. Path Loss Model One of the most important propagation characteristics considering its impact on effective usable distance, especially in mmWave frequency bands, is path loss. The feasibility of mmWave band for mobile cellular has been fiercely debated because considerable path loss was expected at mmWave [4]. Additionally, to investigate the feasibility of its coexistence with other wireless networks such as satellite links, the need for an interference analysis has been established using path loss models [26], [27]. In this section, we present path loss models based on our 28 GHz and 38 GHz measurements, compare them with existing lower frequency models, and discuss system impacts. Path loss follows log-log dependency on both TX–RX separation distance and operating frequency [2], [28]. Depending on the number of free variables in linear6) fitting in log-log domain, path loss has been modeled with either the alpha–beta–gamma (ABG) model or the close-in free space (CI) model [29]. The ABG model can be expressed as follows: (1) where f and d are the operating frequency and TX–RX separation distance, respectively. The parameters α, β, and γ are fitting variables that denote distance-dependency coefficient, offset, and frequency-dependency coefficient, respectively. The CI model is given as (2) where n denotes the path loss exponent (PLE), and FSPL(f, 1 m) is the 1 m-reference free-space path loss value at frequency f. The 1 m distance free-space path loss value is typically used for mmWave bands [29]. (3) where c is the speed of light (8 m/s). If frequency dependency is not considered, the ABG model has two parameters (slope α and intercept β), and the CI model has one parameter (slope n). Namely, the ABG model has more degrees of freedom in fitting. In many cellular network standards such as 3GPP [13], WINNER II [30], and ITU-R M.2135 [14], the ABG model was adopted for system design/planning and verification/evaluations. On the other hand, COST-231 [31] and ITU-R P.1411 [32] adopted the CI model. For mmWave bands, the pros and cons of the two models have been widely discussed. Although the ABG model generally provides more accurate predictions, physical interpretations cannot be easily assigned to the variables because it has additional fitting variables such as β and γ in (1). The CI model, however, has only one fitting variable, the path loss exponent n, which provides some insight when compared with free space propagation (). Table 3 shows the ABG and CI model parameters by fitting our 28 GHz and 38 GHz measurement data, in which σ denotes the standard deviation of fitting errors. It should be noted that the table shows the mmWave model parameters and the sub-6 GHz model parameters from 3GPP [13] and ITU-R M.2135 [14] that are valid between 2 GHz and 6 GHz. As reported in [21], [29], [33], we can also see that in the CI model, n is nearly constant for frequency variations. For LOS, n is slightly higher than 2 (the free-space path loss exponent) in outdoor UMi environments but slightly lower than 2 in indoor InH environments. This difference in behavior can be understood if we consider the presence of different construction-related elements and materials in indoor environments that act as local scatterers or reflectors. This behavior can be also observed in NLOS so that n of InH is lower than that of UMi. In the case of ABG parameters, we cannot easily see the deviation from the sub-6 GHz model because α, β, and γ should not be interpreted separately. With these new 28 GHz–38 GHz ABG parameters, we can utilize the sub-6 GHz system design frameworks, such as computer simulation setups. It should be noted that the significance of this table in this study is that the model parameters are obtained with actual omnidirectional antenna measurements (250–400 samples) in both outdoor and indoor environments. Owing to hardware limitations, path loss has been reported in mmWave bands using directional antennas whose data has been processed to allow for synthesis of an omnidirectional antenna [5], [21], [29], [33], [34]. In order to show detailed measurement results, we plot our 28 GHz and 38 GHz path loss measurements and their fitting to the CI model, as shown in Fig. 5. The x-axis denotes the TX–RX 3D distance separation in logarithmic scale. Because 28 GHz and 38 GHz measurement samples were collected at identical TX–RX locations, both 28 GHz and 38 GHz path loss data exhibit similar patterns for both LOS and NLOS, except constant offsets. These offsets are caused by the frequency dependency on path loss and can be calculated by (3): . Figure 5Open in figure viewerPowerPoint Path loss measurement and CI model fitting: (a) UMi and (b) InH. By simply comparing path loss exponents (n), we cannot determine mmWave propagation loss behaviors because the constant offset, calculated using (3), is considerable for mmWave frequency bands. Table 4 shows the offset at 2 GHz, 28 GHz, and 38 GHz (globally, the existing 3G/4G cellular networks are deployed at approximately 2 GHz). When compared with a 2 GHz cellular network, 28 GHz and 38 GHz networks have additional 22.92 dB and 25.58 dB losses, respectively. In order to avoid outages, this loss must be compensated either by adopting a high-gain directional antenna/beamforming or by shrinking cell coverage areas. Apart from the constant path loss offsets, both LOS and NLOS of 28 GHz and 38 GHz measurements do not deviate from lower frequency models, considering the ranges of the path loss exponent, as shown in Table 3. Consequently, our measurements disprove the myth that mmWave NLOS propagation is significantly worse than LOS. The degradation in mmWave propagation affects both LOS and NLOS by the same amount. This is the constant offset, as calculated by (3). From the perspective of mmWave system design, the most important consideration, regardless of LOS or NLOS conditions, is to compensate for constant offset values. Table 3. Path loss model parameters. Environment Link 28 GHz CI 38 GHz CI 28 GHz–38 GHz ABG Existing sub-6 GHz model [13], [14] n σ n σ α β γ σ α β γ σ UMi LOS 2.1 1.8 dB 2.1 1.7 dB 1.9 40.6 dB 1.8 1.7 dB 2.2 28.0 dB 2.0 3.0 dB NLOS 3.0 3.8 dB 3.0 5.4 dB 4.5 3.1 dB 2.0 4.3 dB 3.67 22.7 dB 2.6 4.0 dB InH LOS 1.8 1.4 dB 1.9 1.6 dB 1.8 20.7 dB 2.9 1.5 dB 1.69 32.8 dB 2.0 3.0 dB NLOS 2.5 6.4 dB 2.6 2.5 dB 1.5 29.5 dB 3.6 5.5 dB 4.33 11.5 dB 2.0 4.0 dB Table 4. Examples of constant offset calculated by free-space path loss (FSPL). 2 GHz 28 GHz 38 GHz FSPL 38.46 dB 61.38 dB 64.04 dB IV. Multipath Characteristics Thus far, we have discussed path loss characteristics, which from the perspective of system design, can be translated into received power/SNR statistics. Along with this holistic view of received signal components, this section focuses on component-level analyses of received signals in time (delay) and spatial (angle) domains. The delay analysis will provide frequency selectivity information of mmWave channels along with system bandwidth and symbol duration period. The angular analysis can be used to determine the beamwidth in beamforming [35], [36]. 1. SAGE for Rotating Directional Scanned Data To extract individual MPCs from the measurements, we applied the SAGE algorithm [15]. Although SAGE was developed to process array antenna data, recent studies have focused on the use of SAGE for directional scanning measurement data, especially in mmWave frequency bands [16], [37]. This algorithm is employed on the assumption that the received signal in each direction corresponds to a received signal of an antenna element in a virtual circular array, thus indicating that the channel is stationary during antenna rotation because the received signals are gathered by sequential scans. Specifically, the received signal captured at the ith angle direction can be expressed as a sum of MPCs as follows: (4) where sl is the lth MPC, and L and t denote the total number of MPCs and time index, respectively. Each MPC can be expressed as (5) where αl, Ωl, and τl are the gain, angle of arrival (AoA), and delay of the lth MPC, respectively. The transmitted signal and additive noise are denoted by x and w, respectively. It should be noted that the difference from the regular SAGE [15] can be found in c(i)(Ωl) because this quantity takes into account the antenna shaping factor (or beam pattern) for the arrival direction of the lth MPC. Therefore, with the models in (4) and (5), we can easily apply the SAGE algorithm. After several trial-and-error attempts and considering environmental structures, we set the maximum number of MPCs for SAGE to be 100 and 200 for LOS and NLOS, respectively, which effectively captures the most channel power. Figure 6 shows our SAGE run: the power azimuth-delay spectrum is nearly identical to the reconstructed spectrum with the channel impulse responses (CIRs) from the SAGE because the residuals are negligible. Figure 6Open in figure viewerPowerPoint An example of SAGE results in a 38 GHz NLOS InH environment: Power azimuth angle-delay spectra calculations: (a) spectrum with original CIRs, (b) reconstructed spectrum from SAGE results, and (c) residual between (a) and (b). With the resolved MPCs from the SAGE runs, we can analyze the temporal and spatial dispersion characteristics. The delay spread is the time difference between the first significant MPC and the last MPC, in which we determine the significance by power level difference from the strongest MPC by 25 dB, as recommended in [32], [38]. The rms delay spread is calculated from the second-order central moment [30] as follows: (6) where and Further, p(·) is the power distribution function, and P(·) is the power delay profile. Similarly, rms angular spread can be calculated by (7) where PAS(Ω) is the power angular spectrum [30]. Figure 7 shows the cumulative distribution functions (CDFs) of delay and angular spreads for our UMi and InH measurements in both 28/38 GHz LOS/NLOS environments, and Table 5 shows their 10%, 50%, and 95% values of dispersion statistics. Figure 7Open in figure viewerPowerPoint Multipath dispersion statistics. Table 5. Typical rms delay and angular spread values. Link type CDF percentile Delay spread (ns) Angular spread (°) 28 GHz 38 GHz 28 GHz 38 GHz UMi LOS 10% 2.2 3.5 7.8 8.6 50% 10.8 8.4 16.8 18.2 95% 29.2 37.8 47.1 41.5 NLOS 10% 12.6 13.4 22.0 24.2 50% 44.6 40.0 42.2 38.7 95% 99.4 106.6 77.7 71.0 InH LOS 10% 12.3 8.3 10.0 11.0 50% 42.5 32.6 30.6 36.2 95% 87.9 92.9 74.3 79.6 NLOS 10% 37.1 43.2 40.4 46.2 50% 108.3 79.3 73.9 67.3 95% 229.5 205.1 91.1 88.9 2. Multipath Cluster Characteristics From a system design perspective, cluster-level analysis by grouping similar MPCs is crucial to improve channel reliability by utilizing diversity properties. Typical standard models have been proposed with cluster parameters as functions of scenario environment and operating carrier frequency [12], [30]. In this section, we present multipath cluster characteristics on the basis of our SAGE results calculated with our 28 GHz and 38 GHz UMi and InH measurements. For clustering, we applied the K-power-means algorithm [39] that calculates the distance by weighting with the power of the MPCs. Figure 8 shows a clustering result of the SAGE-output MPCs. Depending on the distances in the delay-AoA dimension, MPCs are clustered into different groups. Figure 8Open in figure viewerPowerPoint A clustering example of SAGE results in a 38 GHz NLOS InH environment: (a) MPCs from SAGE and (b) clustering result. Table 6 shows the 10%, 50%, and 95% CDF values of cluster-level statistics. When compared with the sub-6 GHz models [12], [30], fewer clusters are present, thus indicating that lower diversity gain can be achieved in mmWave channels. Figure 9 shows the CDF of cluster numbers, cluster delay spread, and cluster angular spread of our measurements. As can be seen, InH has a greater number of clusters than UMi, thus leading to larger delay and angular spreads. Table 6. Typical values of cluster number, delay spread, and angular spread. Link type CDF percentile No. of clusters Cluster DS (ns) Cluster AS (°) 28 GHz 38 GHz 28 GHz 38 GHz 28 GHz 38 GHz UMi LOS 10% 1.0 1.0 0.0 0.0 2.1 2.3 50% 2.2 2.3 1.1 1.3 5.0 5.0 95% 5.6 5.4 3.7 6.1 16.3 14.7 NLOS 10% 3.7 3.5 1.5 0.9 3.8 3.3 50% 5.9 6.6 8.3 6.3 10.5 8.6 95% 9.7 10.6 29.8 21.5 27.2 25.8 InH LOS 10% 1.0 1.2 0.3 0.4 1.1 1.9 50% 3.7 4.5 1.7 1.5 5.0 4.9 95% 10.3 8.4 18.8 16.1 24.3 21.3 NLOS 10% 2.9 5.4 5.0 9.2 6.2 10.5 50% 8.8 8.5 17.0 15.6 15.3 17.6 95% 12.8 12.4 71.4 55.8 29.5 34.8 Figure 9Open in figure viewerPowerPoint Cluster-level MPC statistics. Along with the clustering results, we generated 3GPP-like stochastic channel model parameters, as shown in Table 7. We followed the 3GPP notations and symbol conventions used in [12], [30]. Therefore, μ and σ denote the average mean value and standard deviation of the associated channel parameter, respectively. The table also shows comparisons with the sub-6 GHz models, which will eventually provide information on how sub-6 GHz systems must be modified to be successfully operated in mmWave bands. Table 7. Channel model parameters for UMi & InH environments. Parameters UMi InH 28 GHz 38 GHz Sub-6 GHz 28 GHz 38 GHz Sub-6 GHz LOS NLOS LOS NLOS LOS NLOS LOS NLOS LOS NLOS LOS NLOS Delay spread (DS) log10 (s) μ −8.1 −7.5 −8.0 −7.4 −7.19 −6.89 −7.4 −7.0 −7.5 −7.1 −7.7 −7.41 σ 0.4 0.4 0.4 0.3 0.4 0.54 0.3 0.3 0.4 0.2 0.18 0.14 AoA spread (ASA) log10 (°) μ 1.3 1.6 1.3 1.6 1.75 1.84 1.5 1.9 1.6 1.8 1.62 1.77 σ 0.3 0.2 0.2 0.2 0.19 0.15 0.3 0.2 0.3 0.1 0.22 0.16 K-factor (dB) μ 11 N/A 9 N/A 9 N/A 9.6 N/A 7.8 N/A 7 N/A σ 2 N/A 1 N/A 5 N/A 4.9 N/A 3.6 N/A 4 N/A Cross-correlation ASA vs. DS 0.6 −0.2 0.8 0.2 0.8 0.4 0.3 0.0 −0.2 0.3 0.8 0 ASA vs. SF 0 0.1 −0.4 0.2 −0.4 −0.4 −0.8 −0.5 −0.2 0.1 −0.5 −0.4 DS vs. SF −0.5 −0.2 −0.2 −0.3 −0.4 −0.7 −0.1 −0.1 0.1 0.0 −0.8 −0.5 ASA vs. K −0.2 N/A 0.1 N/A −0.3 N/A −0.9 N/A 0.0 N/A 0 N/A DS vs. K −0.4 N/A −0.4 N/A −0.7 N/A −0.2 N/A −0.5 N/A −0.5 N/A SF vs. K −0.1 N/A −0.2 N/